Real-time applications, such as ticket booking systems, video streaming systems, social networking services, and online games, are prevalent nowadays. Low latency between a user and a server is required in such applications. A distributed server approach is used instead of a single server one to achieve this. In such a situation, selecting which users connect to which servers significantly determines the system’s performance. There is a problem with the distributed server approach; each server may fail due to maintenance or breakdown. Single-server failure is more popular than multiple-server failure, so this paper focuses on single-server failure. In this paper, we assume that the capacity of each server is variable and minimize the total capacity of the servers. We also set the assumption that each user is allowed to connect exactly two servers. Under this assumption, we select each user’s main and backup servers. This paper proposes two approximation algorithms for a server allocation problem to minimize total capacity under server failure. We formulate the problem as an integer linear programming (ILP) problem. We show that the proposed algorithms have an approximation rate of$4/3$. One of them is an improved version of the other one. The algorithms are based on a 2-approximation algorithm of a vertex cover problem, in which taking a maximal matching is the key idea. Numerical results show that the proposed algorithms run significantly faster than ILP. The performance of the proposed algorithms depends on the topology; still, in randomly generated general networks, it increases the total capacity by 2.03% or less than ILP on average.

Spectrally-spatially elastic optical networks (SS-EONs) enable ultra-high data rate transmission, which raises critical concerns about physical-layer security vulnerabilities, particularly against eavesdropping and unauthorized network access. Dynamic resource allocation through lightpath reconfiguration presents an effective approach to improving security by reducing request exposure windows. However, implementing secure reconfiguration in SS-EONs introduces significant complexity due to the complex relationships between spectral allocation and spatial resource management constraints. This paper proposes an analytical model for blocking performance with security-aware reconfiguration (AnalyticalSAR) in SS-EONs based on continuous-time Markov chain analysis to tackle these security challenges. The AnalyticalSAR provides analytical assessment of how spectrum reconfiguration affects both network security and blocking performance while accounting for inter-core and intermode crosstalks. The model generates all viable states accounting for spectrum reconfiguration processes and their corresponding transitions to establish state probabilities. Our analysis incorporates two distinct spectrum allocation policies: core-modespectrum random fit (CMS-RF) and core-mode-spectrum first fit (CMS-FF) policies. Our model supports diverse traffic scenarios, including single-class requests with uniform slot requirements and multi-class requests including heterogeneous bandwidth demands. To overcome computational complexity limitations in single-hop analyses, we develop an heuristic iterative approach and subsequently extend this approach to multi-hop network scenarios. We compare AnalyticalSAR, the heuristic iterative approach, and Monte Carlo simulation studies for a single-hop link. Analytical evaluation reveals that random spectrum reconfiguration substantially improves security metrics while introducing minimal blocking probability increases. These performance trade-offs depend critically on number of spectrum reconfiguration, link and network load conditions, and available link capacity. The results validate that AnalyticalSAR achieves an effective compromise between security enhancement and operational performance, providing a practical framework for secure resource management in SS-EON deployments.

Low-rate Denial of Service (LDoS) attacks use short, regular bursts of traffic to exploit vulnerabilities in network protocols. They are a major threat to network security, especially in Software-Defined Networking (SDN) frameworks. These attacks are challenging to detect and mitigate because of their low traffic volume, making it impossible to distinguish them from normal traffic. We propose a real-time LDoS attack detection and mitigation framework that can protect SDN. The framework incorporates a detection module that uses a deep learning model, such as a Generative Adversarial Network (GAN), to identify the attack. An efficient mitigation module follows detection, employing mechanisms to identify and filter harmful flows in real time. Deploying the framework into SDN controllers guarantees compliance with OpenFlow standards, thereby avoiding the necessity for additional hardware. Experimental results demonstrate that the proposed system achieves a detection accuracy of over 99.98% with an average response time of 8.58 s, significantly outperforming traditional LDoS detection approaches. This study presents a scalable, real-time methodology to enhance SDN resilience against LDoS attacks.

In an Internet Protocol network running a link-state routing protocol, determining the link weights selects each source-to-destination route on which the sum of the link weights is minimized. Previous studies have focused on determining physical link weights to reduce the network congestion ratio in case of physical-link failures. However, no study has yet addressed a model that determines link weights by incorporating traffic-engineering (TE) links and investigates the effect of incorporating TE links on reducing network congestion. The link-state routing protocol treats a TE link as a logical, direct link between nonadjacent nodes. This paper proposes a link-weight design model with TE links based on preventive start-time optimization (PSO) for handling single physical-link failures, called PSO-TE. The model considers physical and TE links when determining the link weights under the assumption of all single physical-link failures. It identifies the set of link weights that minimizes the worst-case network congestion ratio across all considered failure patterns. Introducing TE links does not require additional physical link resources and thus does not increase capital expenditures. Numerical results demonstrate that PSO-TE reduces the worst-case network congestion ratio compared to PSO without TE links. PSO-TE reduces the worst-case network congestion ratio compared with other models, including PSO without TE links, start-time optimization, and inverse capacity weighting.

Real-time applications require low latency and strict event ordering to ensure seamless operation. Distributed server processing is effective for this purpose, and there are two synchronization algorithms: a conservative synchronization algorithm (CSA) and an optimistic synchronization algorithm (OSA). OSA improves delay performance compared to CSA. While prior studies have considered OSA, they have not incorporated the impact of server failures. This paper proposes an OSA-based server allocation model for delay-sensitive applications with preventive start-time optimization (PreSO) under single-server failures (OSA-PreSO). The proposed OSA-PreSO model minimizes the largest total delay across all failure scenarios while satisfying constraints in OSA with PreSO under single-server failures. We formulate the proposed model as an integer linear programming (ILP) problem. In OSA-PreSO, the objective is to minimize the largest total delay across all failure scenarios, without giving special consideration to the total delay in the no-failure scenario. As a result, a penalty arises in the form of an increased total delay in the no-failure scenario. To reduce the penalty, we develop an improved OSA-PreSO model, OSA-PreSO-LP (low-penalty), which reduces the total delay in the nofailure scenario while maintaining the same delay characteristics in failure scenarios. We prove that the decision version of OSA-PreSO is NP-complete. We introduce heuristic algorithms to handle large-scale problems. Numerical results show that the proposed OSA-PreSO model reduces the delay compared to the conventional CSA-based model by effectively utilizing server memory resources. We observe that the proposed model achieves a lower largest total delay than start-time optimization and provides greater stability by preventing unnecessary user reassignments compared to run-time optimization. Numerical results also show that OSA-PreSO-LP reduces the penalty at most by 83%, while maintaining the same delay characteristics in failure scenarios.

Real-time applications require low latency and event order guarantees. Distributed server processing is effective for this purpose, and data consistency between servers is crucial. Although existing models in previous work handle data consistency, they do not address server failures. This paper proposes a server allocation model for a consistency-aware multi-server network for delay-sensitive applications with preventive start-time optimization (PSO) under single-server failures. The proposed model considers data consistency between servers and handles single-server failures with PSO. PSO determines the assignment to minimize the worst-case delay over all possible failure scenarios while avoiding service disruption for users connected to non-failed servers. We formulate the proposed model as an integer linear programming (ILP) problem. The decision version of the server allocation problem is proven to be NP-complete, and it becomes difficult to solve in a practical time when the problem size is large. We develop two polynomial-time approximation algorithms with theoretical performance analysis. Numerical results show that the proposed model outperforms start-time optimization in terms of the largest total delay and run-time optimization in terms of avoiding instability. The results also show that the faster of our two developed algorithms achieves a speedup ranging from 2.26×103to 4.37×106times compared to the ILP approach, while the maximum delay is, on average, only 1.029 times the optimal value. The results indicate that the speedup effect becomes more significant as the number of users and servers increases.

This Special Issue contains a set of remarkable papers covering various recent research advances towards resilient Communication Networks for an hyper-connected World. Papers are organized into five categories: (i) Resilient Architectures for Next-Generation Networks, (ii) Edge, IoT, and Cyber-Physical Systems, (iii) Vehicular, Mobile, and Aerial Networks, (iv) Optical, Hybrid, and Satellite-based Resilient Communications, and (v) Security, Trust, and Resilience in Services and Applications. The editorial begins with an overview of the field and proceeds with a summary of the twenty-two papers included in this Special Issue.

Middleboxes play a critical role in network operations, providing various network service functions, and can be implemented as software on general-purpose servers through network function virtualization technology. The unavailability of middlebox functions is an essential metric of the overall quality of network services. This paper proposes an analytical model that calculates the unavailability of middlebox functions, where multiple backup servers protect one or more functions, and one backup server can recover at most two functions simultaneously, which we call the double-capacity multiple-backup model. While the single-capacity multiple-backup model, where each backup server can recover at most one function, has been studied, the double-capacity multiple-backup model has not been addressed. The proposed double-capacity multiple-backup model allows for load balancing among backup servers. The proposed model can have two different workload strategies: load-persistent (LP) and load-distributing (LB). We use a Markov chain to analyze state transitions and develop equilibrium-state equations, providing a method to compute function unavailability for each strategy. Numerical results observe that these two strategies have almost the same unavailability. Since the LB strategy incurs additional operational costs compared to the LP strategy, the LP strategy is preferable. We also find that the double-capacity multiple-backup model reduces unavailability by 7.21–81.9% compared to the single-backup model in the examined cases.

Optical circuit switching provides high-capacity and low-latency data transmission capabilities. A Clos network is well known as a class of non-blocking switching networks. A prior study addressed designing a three-stage twisted and folded Clos network (3TF), which has a larger switching network size, i.e., the number of terminals connected to the network, than its two-stage variants. That study used a sufficient condition for strict-sense non-blocking (SNB) in 3TF to solve an optimization design problem. However, this condition may be stricter than necessary to guarantee strict-sense non-blocking, suggesting that by using the necessary and sufficient condition, the constraints can likely be relaxed, allowing for configurations with a larger switching network size. No study has addressed the necessary and sufficient condition for SNB in the whole 3TF. A past study derived the SNB condition for the three-stage folded Clos (3F), the basis of 3TF, assuming that connections within the same first- and second-stage group loop back no further than the second stage. Its necessary and sufficient condition in the general case, however, remains unexamined, despite an initial study on the SNB condition for 3F. This paper proposes a design model of 3TF based on the necessary and sufficient condition for strict-sense non-blocking. We derive and prove the necessary and sufficient condition for strict-sense non-blocking in the whole 3TF. The proposed design model considers two cases: one where no blocking is allowed and another where the blocking probability does not exceed a given admissible threshold, based on our derived necessary and sufficient condition for strict-sense non-blocking in 3TF. We formulate the problem based on the necessary and sufficient condition to obtain the configuration of the proposed model that maximizes the switching network size. We theoretically show that the maximum switching network size obtained by solving the proposed model is always larger than or equal to that obtained by solving the existing 3TF design model. Numerical results show that the proposed model achieves a larger switching network size than the existing 3TF design model.

Signal transmission using counter-propagation is increasingly adopted to enhance resource utilization in spectrally-spatially elastic optical networks (SS-EONs). Crosstalk is a well-known drawback of SS-EONs that increases blocking probability. Evaluating blocking probability analytically in counter-propagation is challenging due to additional constraints. Current studies typically employ simulation-based techniques or do not consider dynamic scenarios in their analytical models to estimate blocking probability. This paper proposes AnalyticalCP, an exact analytical continuous-time Markov chain model (CTMC) for SS-EONs considering counter-propagation, which estimates blocking probabilities and enhances resource utilization. AnalyticalCP is constructed in two phases: (i) First, a bipartite graph is formed to generate two separate sets of cores and modes, using an optimization approach focused on minimizing interference among cores and modes. (ii) Second, the resultant sets of cores and modes serve as input parameters for the analytical model, streamlining the generation of all feasible states and transitions to estimate exact blocking probability. AnalyticalCP accommodates both single- and multi-class requests: single-class requests uniformly utilize spectrum slots, while multi-class requests adapt slot usage to match bandwidth requirements. Numerical results indicate that AnalyticalCP and the simulations yield comparable performances in terms of blocking probabilities and resource utilization, with the iterative approximate model also demonstrating acceptable accuracy. AnalyticalCP outperforms a benchmark model that considers vertex elimination during bi-partitioning. We extend our CTMC model with a Markov-modulated Poisson process (MMPP) to evaluate bursty-traffic behavior in SS-EONS.

Service function chains (SFCs) that utilize programmable data plane (PDP) have attracted attention due to their ability to balance high packet processing performance and flexibility. To adapt to network fluctuations, SFCs require periodic updates; paths are migrated from old to new ones. Ensuring state consistency during such migrations is essential for maintaining accurate packet processing. This paper presents an experimental analysis of migration time for an SFC with PDP-based network functions (NFs) by implementing two network systems. The first system, called System A, employs individual switch controllers, whereas the second system, called System B, consolidates them into the aggregated switch controller to eliminate overhead of aggregating state. An SFC migration time is measured under various conditions for both systems. When a firewall NF with 1000 table entries is used, the state extraction and conversion time accounts for 70% of the total migration time in System A. System B removes state conversion by integrating switch controllers and reduces the migration time by 76% compared to System A. In System B, state writing accounts for 72%, state extraction for 21%, and command propagation for 6.5% of the total migration time.

Spectrally-spatially elastic optical networks (SS-EONs) face challenges such as fragmentation and crosstalk, which lead to a rise in blocking probability. Accurate modeling of fragmentation and blocking probability is complex due to multiple constraints in SS-EONs. We evaluate the average fragmentation without considering spectrum contiguity constraints while simultaneously avoiding inter-core and inter-mode crosstalks in SS-EONs, which has not been studied in previous research. This study employs a continuous-time Markov chain-based analytical model to compute average fragmentation and blocking probability under different spectrum allocation policies—core-mode-spectrum random fit (CMS-RF) and core-mode-spectrum first fit (CMS-FF)—without contiguity constraints while simultaneously avoiding inter-core and inter-mode crosstalks in SS-EONs. The model considers both single-class and multi-class requests, where single-class requests use a uniform number of slots, and multi-class requests have variable slot requirements. Additionally, an iterative approximation model is introduced to improve scalability in single-hop links. Numerical results show that both the analytical model and simulations demonstrate lower blocking probability when contiguity constraints are not enforced, with CMS-FF performing better than CMS-RF in reducing fragmentation.

RFC 6437 introduces the 20-bit flow label field in the IPv6 header and recommends it to be used alongside source and destination IP addresses for efficient flow classification. A recent paper from Google authors explains how they have begun to use this field in their IPv6 network to select an alternative path, thereby preventing congestion or outages. However, there is no prior research investigating whether connections make use of this field on the Internet. In this paper, we present a comprehensive analysis of the 9-year MAWI dataset (2016-2024), the Internet traces gathered from a large backbone link in Japan, and examine whether the IPv6 flow label is indeed employed on the Internet. Our findings reveal an increasing upward trend in the adoption of the flow label field. In particular, this trend correlates with the use of QUIC, whose specification—different from TCP—explicitly repeats the recommendation to utilize the flow label.

Optical circuit switching (OCS) is becoming used in some data center networks due to its low power consumption, low latency, and high bandwidth. Previous research introduced a design model for a twisted and folded Clos network (TF-Clos) as a data center network to maximize the switching network size, i.e., the number of connected terminals, while guaranteeing the admissible blocking probability. The previous model assumes that request active rates from all the terminals are identical. However, it is an overly conservative design when the active rates differ, resulting in a smaller switching network size than desired. This paper proposes a terminal-shuffling (TS) scheme for designing an OCS TF-Clos network with an admissible blocking probability guarantee, which supports different active rates. Each terminal can arbitrarily choose any leaf switch to connect, making the network design more adaptable to varying conditions. A patch panel or direct termination by operators can wire optical fibers between the terminals and the leaf switches. We formulate a TS-based TF-Clos design problem to maximize the switching network size. We develop an approximation approach to find a feasible solution to the optimization problem. Numerical results demonstrate that the switching network size of the proposed TS scheme is larger than that of baseline schemes.

The introduction of optical-circuit-switches (OCSes) has enabled the implementation of capacity- and energy-efficient networks in production datacenters. To correctly operate optical-circuit-switched datacenter networks (OCS DCNs), fibers between pairs of terminals (e.g., servers or top-of-rack switches) and OCSes should be verified before starting operations. However, this task is difficult because OCSes cannot use topology discovery or link monitoring functions, which are only available on electrical packet switches. Motivated by this challenge, we investigated a fiber topology and quality verification (FTQV) problem for OCS DCNs in this paper. Though a previous study inspected fibers in hierarchical OCS DCNs using only one dedicated tester for fiber probing, making the process time-consuming, we consider using digital diagnostic monitoring (DDM) functions at multiple transceivers for fiber inspection. We thus developed solid theories for correctly and quickly inspecting fibers even when multiple probes are sent in parallel. We also developed an algorithm that correctly and quickly solves the FTQV problem on the basis of our theories. Numerical experiments showed that our algorithm completes FTQV at most 48.7 times faster than a baseline algorithm.

Real-time applications require low latency and event order guarantees. While distributed server processing is effective, data consistency across servers is crucial. Existing models address consistency but overlook server failures. This paper proposes a server allocation model for a consistency-aware multi-server network for delay-sensitive applications with preventive start-time optimization (PSO) under single-server failures. PSO determines the assignment to minimize the worst-case delay over all possible failure scenarios while avoiding service disruption for users connected to non-failed servers. We formulate the proposed model as an integer linear programming (ILP) problem. The decision version of the server allocation problem is proven to be NP-complete, and it becomes difficult to solve in a practical time when the problem size is large. We develop a polynomial-time approximation algorithm with theoretical performance analysis. Numerical results show that the proposed model outperforms start-time optimization in terms of the largest total delay and run-time optimization in terms of avoiding instability. Numerical results also show that our developed algorithm achieves a maximum speedup of 33.8 times compared to the ILP approach, while the maximum delay is, on average, only 1.016 times the optimal value.

Data center networks facilitate large-scale data processing by interconnecting multiple switching devices. Optical circuit switching (OCS) provides high transmission capacity and energy efficiency. It establishes dedicated paths for data transfer, ensuring reliable communication. A Clos network is widely used among multi-stage switching architectures due to its scalability and structured design. This paper investigates models for designing folded/unfolded Clos networks with an admissible blocking probability to maximize OCS network size. While previous studies have examined fundamental and stage-extended Clos networks, they have not addressed unfolded Clos network structures that maintain an admissible blocking probability across different configurations. To fill this gap, we introduce unfolded Clos network structures that ensure an admissible blocking probability for both fundamental and extended stages. We also discuss connection admission control mechanisms tailored to these network models. A key focus of this study is a comprehensive performance evaluation, including switching network size and computation time. Furthermore, we explore an alternative approach employing multiple network planes to enhance scalability and flexibility. The findings provide valuable insights into the design of large-scale OCS networks with controlled blocking probabilities.

The exploration of virtualization in spectrally-spatially elastic optical networks (SS-EONs), particularly focusing on virtual optical network embedding (VONE), is an emerging technology to enhance resource utilization and transport capacity. However, managing both inter-core crosstalk (IC-XT) and inter-mode crosstalk (IM-XT) in virtualized SS-EONs poses significant challenges. To tackle this, the paper proposes an optimization model for VONE in SS-EONs, named VneXT-Aw, which incorporates an XT-aware approach. VneXT-Aw employs node and link mapping techniques to seamlessly integrate VON requests into the substrate SS-EONs, ensuring precise node mapping and capacity adherence. It allocates spectrum slots along routing paths in the substrate network, ensuring spectrum contiguity, continuity, and mode continuity for virtual links, all while considering the XT-aware approach. The optimization problem for VneXT-Aw is formulated as a mixed-integer linear programming (MILP) problem. Recognizing the complexity of MILP for larger instances, the paper introduces two heuristic approaches: MILP-based heuristic (MILP-h) and rank-assisted simulated annealing (Rasa-h). A comparative analysis reveals that MILP-h accommodates more requests than Rasa-h but sacrifices computational efficiency. This trade-off highlights the delicate balance between request accommodation and computational complexity, providing insights into the practical implementation of VON embedding in SS-EONs. Furthermore, VneXT-Aw outperforms the benchmark scheme by utilizing the XT-avoided approach.

Internet-of-Things (IoT) services require high performance regarding low delay and fault tolerance. Distributed server allocation is well-suited for meeting these requirements in IoT monitoring services. Previous work focused on reducing delay but overlooked the need for fault tolerance in distributed server allocation. This paper proposes a distributed server allocation model based on preventive start-time optimization (PSO) for IoT monitoring services against server failure. The proposed model preventively determines the server allocation to minimize the largest maximum delay between IoT devices and application servers and between database and application servers among all failure patterns. We formulate the proposed model as an integer linear programming (ILP) problem. We introduce a server allocation algorithm based on PSO to accelerate the computation to obtain an optimal server allocation, compared to the ILP approach. We prove that the introduced algorithm obtains a PSO-based optimal allocation in polynomial time. Numerical results show that the introduced algorithm outputs an optimal server allocation faster than the ILP approach. We compare the PSO-based server allocation with allocations based on the start-time and run-time optimization. We observe that the PSO-based allocation reduces the largest maximum delay by 5.5% for a network model with eleven servers compared to the start-time optimization and avoids unnecessary network disconnections while increasing the maximum delay by 5.1% compared to the run-time optimization.

Delay-sensitive applications have been provided through a low-delay network utilizing multiple edge clouds. For applications that involve sharing status among multiple users, it is crucial to prevent longer communication delays for users who are farther from the application server compared to those who are closer. To address this issue, this paper proposes a distributed processing network design scheme for virtual processing platforms using low-delay networks and widely distributed servers. The proposed scheme introduces Tapl as a given parameter for correcting events in occurrence order. Events within Tapl delay are sorted in occurrence order. The proposed scheme can change its operation mode from a conservative synchronization to an optimistic synchronization depending on the setting of Tapl. The proposed scheme is formulated as a mixed-integer linear programming problem to determine users’ and servers’ distributed processing network configuration. We evaluate the proposed scheme on two different network topologies. Numerical results indicate that, depending on the setting of Tapl, the proposed scheme can reduce the maximum amount of memory used for rollback processes in optimistic synchronization-based applications or realize a conservative synchronization algorithm. The computation time under the condition of 1000 users is within a maximum of nine sec, an acceptable amount of time for preparation before starting a planned service. These results indicate that the proposed scheme realizes event order correction with excellent delay characteristics and applies to virtual processing platforms.

Future advancements in data centers are anticipated to incorporate advanced circuit switching technologies, especially optical switching, which achieve high transmission capacity and energy efficiency. Previous studies addressed a Clos-network design problem to guarantee an admissible blocking probability to maximize the switching capacity, which is defined by the number of terminals connected to the network. However, as the number of available${N} \times {N}$switches increases, the switching capacity no longer increases due to the switch port limitation. This paper proposes a design of a multiple-plane twisted-folded (TF) Clos network, named MP-TF, to enhance the switching capacity, which is limited by the original TF-Clos, by guaranteeing an admissible blocking probability. MP-TF consists of identical M TF-Clos planes and pairs of a$1\times {M}$selector and an${M} \times 1$selector, each pair of which is associated with a transmitter and receiver pair. We formulate a design model of MP-TF as an optimization problem to maximize the switching capacity. We introduce connection admission control in MP-TF, named MP-CAC. We derive the theorem that the MP-TF design model using MP-CAC guarantees the admissible blocking probability. Numerical results observe that MP-TF increases the switching capacity as the number of TF-Clos planes when available${N} \times {N}$switches are sufficient; for example, with seven planes, the switching capacity is 1.97 times larger than that of one plane, given a request active probability of 0.6 and an admissible blocking probability of 0.01. We find that the computation time for MP-TF diminishes with an increase in the number of TF-Clos planes. Designing MP-TF is similar to designing a single TF-Clos plane, differing mainly in the handling of connection admission control. With a larger number of${N} \times {N}$switches, MP-TF enables the design of a smaller TF-Clos plane. We provide the analyses of optical power management and network cost of MP-TF.

In a software-defined network (SDN), routes of packet flows need to be updated in situations such as maintenance and router replacement. Each flow is migrated from its old path to new path. The SDN update has an asynchronous nature; the time when the switches process commands by the controller varies depending on flows. Therefore, it is difficult to control an order of flow migrations, and packets can be lost by congestion. Existing models divide the time axis into rounds and assign migrations to these rounds. However, congestion caused by multiple migrations in the same round is uncontrollable. Based on the probability distribution of time required for each migration, congestion can occur. This paper proposes a flow update model which minimizes the expected amount of excessive traffic by shifting the probability distributions. The time axis is divided into time slots which are fine-grained than rounds, so that each probability distribution is shifted. The proposed model assigns the time when the controller injects a command of flow migration to time slots. The proposed model is formulated as an optimization problem to determine the command times to minimize the expected amount. This paper introduces two methods to compute the expected amount. This paper also introduces a two-stage scheduling scheme (2SS) that divides the optimization problem into two stages. 2SS suppresses the computation time from$\mathcal {O}(|T|^{|F|-1})$to$\mathcal {O}\left ({{|T|^{{}\frac {|F|-1}{2}}}}\right)$at the cost of including at most 0.12% error. 2SS suppresses the amount of excessive traffic than an existing model by at most 71.2%.

In real-time applications under wide-area networks, providing a demanded quality of service for end users is an issue. Recent studies adopt distributed processing for server selection problems to reduce data synchronization delay and total interaction delay, assuming that link delays over the distributed system are exactly known. No study has addressed the problem of such a distributed server selection in properly handling the delay uncertainty. This paper proposes a robust optimization model for the distributed server selection problem against the delay uncertainty. We handle the delay uncertainty of user-server and server-server links by defining two -ellipsoidal uncertainty sets. The proposed model determines allocated servers for multiple users to minimize the weighted sum of data synchronization delay and total interaction delay over the distributed system. We formulate the proposed model as a mixed integer second-order cone programming problem. We prove that the distributed server selection problem with uncertain delays is NP-complete. We compare the proposed model with baseline models, focusing on delay uncertainty and distributed processing. The numerical results show that the proposed model can achieve a lower objective value than the baseline models, indicating the benefit of utilizing -ellipsoidal uncertainty sets to handle delay uncertainty.

In network function virtualization, a network service is provided by a service function chain (SFC), which consists of a chain of virtual network functions (VNFs) within a specific order. SFC parallelism allows parallel processing among VNFs to reduce the end-to-end service delay. Existing works handle the service delay without considering traffic uncertainty, which leads to degraded performance on parallel structure balancing and deployment cost saving in the parallelized SFC deployment problem. This paper proposes a robust deployment model for parallelized SFCs against traffic uncertainty that satisfies the requirement of balanced parallel structures and minimizes the deployment cost. We define a traffic uncertainty set that handles both the variation of service traffic arrival rates and the fluctuation of parallel structures. We apply VNF sharing to improve the efficiency of resource allocation. We formulate the proposed model as a mixed integer second-order cone programming (MISOCP) problem. We introduce a heuristic algorithm to handle larger-size problems, where the MISOCP approach is intractable to obtain a solution in a practical time. Numerical results show the advantages of the proposed model in terms of deployment cost over the baseline models.

In recent years, space division multiplexing (SDM) technology, particularly multi-core and multi-mode fibers (MCMMFs), has been investigated to overcome physical limitations and enhance transport capacity. When SDM is integrated with elastic optical networks (EONs), it gives rise to an emerging technology known as spectrally-spatially elastic optical networks (SS-EONs). However, SS-EONs face significant challenges, such as fragmentation and crosstalk (XT), which increase blocking probability. Defragmentation is widely regarded as the most effective technique to mitigate fragmentation and reduce blocking probability. However, analytically assessing blocking probability with defragmentation is challenging due to the added constraints. Current studies on MCMMF-based SS-EONs typically rely on simulations or overlook defragmentation in their analytical models. This paper proposes AnDefrag, the first exact analytical continuous-time Markov chain model designed to calculate blocking probabilities in SS-EONs, taking into account defragmentation and the XT-avoided approach. AnDefrag generates all possible states and transitions, avoiding inter-core and inter-mode XTs for both single-class and multi-class requests. Single-class requests utilize an equal number of slots, while multi-class requests require different numbers of slots to meet clients’ needs. For cases where AnDefrag is not scalable, we introduce an iterative approximate model for single-hop links, which is extended to multi-hop networks. Numerical evaluations indicate that AnDefrag outperforms a non-defragmentation-aware benchmark model, as demonstrated through comparison with Monte Carlo simulations for a single-hop link.

Low-delay networking and edge computing will enable mission-critical applications to be delivered over wide-area networks. We consider this trend to be the realization that all users can share an application space without feeling any distance difference. We propose a distributed processing scheme that keeps the order of event occurrence regardless of the distance between users and an application server. The proposed scheme can be applied to both optimistic synchronization algorithms (OSA) and conservative synchronization algorithms (CSA). In the proposed scheme, arrival events with a delay within a predefined set time (correction time) are sorted in order of occurrence before application processing. We formulate the proposed scheme as an integer linear programming (ILP) problem. The objective function of ILP consists of the number of users excluded from the delay quality, the amount of memory consumed for a rollback in OSA, and the maximum end-to-end delay. The three parts of the objective function are set weight and the sum of parts with weight is minimized. We evaluate the proposed scheme for 1000 users distributed in two types of network models. Numerical results indicate that the proposed scheme reduces memory consumption compared to that of the conventional OSA scheme. The proposed scheme works as CSA in which all events are sorted in the occurrence order if the correction time is set above the delay for the slowest event to arrive at the server.

In network function virtualization, a network service is provided by a service function chain (SFC), which consists of a chain of virtual network functions (VNFs) within a specific order. SFC parallelism allows parallel processing among VNFs to reduce the end-to-end service delay. Existing works handle the service delay without considering traffic uncertainty, which leads to degraded performance on parallel structure balancing and deployment cost saving in the parallelized SFC deployment problem. This paper proposes a robust deployment model for parallelized SFCs against traffic uncertainty that satisfies the requirement of balanced parallel structures and minimizes the deployment cost. We define a traffic uncertainty set that handles both the variation of service traffic arrival rates and the fluctuation of parallel structures. We apply VNF sharing to improve the efficiency of resource allocation. We formulate the proposed model as a mixed integer second-order cone problem. Numerical results show the advantages of the proposed model in terms of deployment cost over the baseline models.

Future data centers are anticipated to embrace cutting-edge circuit switching technologies, particularly optical switching, renowned for their heightened transmission capacity and energy efficiency. Optical circuit switching guarantees consistent communication quality by establishing dedicated connections for data transmission and maintaining their integrity. Data centers favor Clos-network structures due to their scalability. This paper examines the advancements in designing Clos networks to boost switching capacity while maintaining internal blocking quality. We analyze the characteristics of Clos network design models, providing essentials. We extensively discuss their performances regarding their switching capacities and computation times. Drawing from the review of research progress, we address the challenges in enhancing the switching capacity of Clos networks for future studies.

Cloud computing (CC) and edge computing (EC) enhance the performance of end devices (EDs) with limited computational power by offloading tasks to cloud and edge servers, respectively. Multi-access edge computing (MEC) further advances EC by integrating wireless network resources, thus improving mobile service efficiency. While CC is well-suited for intensive computational tasks, it may face latency issues due to geographical distances. EC and MEC aim to minimize this latency by deploying server resources closer to EDs, but they encounter challenges due to the limited resources of edge servers. Cooperative task offloading emerges as a solution to address the above challenges of optimizing resource allocation across cloud and edge based on task characteristics. Despite numerous research, existing methods often cover only a portion of the networks and servers, leading to sub-optimal task allocation. Therefore, we propose a cooperative task-offloading method for multi-access edge-cloud networks, simultaneously considering server and link resources, base station (BS), and wireless channel allocation. This method improves task-offloading efficiency by utilizing cooperative multi-group multi-agent deep reinforcement learning (CMG-MADRL) with different agent groups for BS and server allocation. Simulations have demonstrated that our method effectively reduces resource utilization and task latency while minimizing constraint violations.

Recently, multi-core and multi-mode fibers (MCMMFs) have been considered to overcome physical limitations and increase transport capacity. They are combined with elastic optical networks (EONs) to form spectrally-spatially elastic optical networks (SS-EONs), an emerging technology. Fragmentation and crosstalk (XT) are well-known drawbacks of SS-EONs that increase blocking probability; evaluating blocking probability analytically is difficult due to additional constraints. When calculating blocking probabilities in MCMMFs-based SS-EONs, it is observed that all current studies either employ simulation-based techniques or do not consider defragmentation of their analytical models. This paper proposes an exact analytical continuous-time Markov chain model for blocking probabilities, named AnalyticalDF, in SS-EONs, which considers defragmentation and the XT-avoided approach. AnalyticalDF generates all possible states and transitions while avoiding inter-core and inter-mode XTs for single-class and multi-class requests. Single-class requests utilize the same number of slots, whereas multi-class requests adopt varying numbers of slots to accommodate client needs. We introduce an iterative approximation model for a single-hop link when AnalyticalDF is not tractable due to scalability. We evaluate AnalyticalDF, the iterative approximate model, and simulation studies for a single-hop link. The numerical results indicate that AnalyticalDF outperforms a non-defragmentation-aware benchmark model.

The domain name system (DNS) is essential to access Internet services. However, although DNS has become a vital part of Internet users in the network and facilitates the process, it is still a critical part of the service chain. This is principally due to its resolver component, which can negatively impact the final user experience of consuming services. Furthermore, to improve service quality and increase user experience, this study intends to mitigate the latency effect of the resolution element of domain name systems in the network. Some studies have developed strategies to reduce response time to resolution. Still, they needed some integrated module in the user’s devices or a more complex and expensive service architecture that was not scalable for edge deployments. This study proposes integrating prediction techniques and a user-oriented resolution name approach to improve the overall quality of access to the Internet for users, even when there are limited computing resources in the edge infrastructure.

This paper proposes a polynomial-time algorithm for a server allocation problem in delay-sensitive Internet-of-Things (IoT) monitoring services. The server allocation problem determines the appropriate servers to which the database and application are allocated to minimize the maximum delay between the latest update of reference data and the start of application processing for monitoring data . The server allocation problem was previously handled by expressing it as an integer linear programming (ILP) problem. Nevertheless, it fails to meet the computational time complexity needed to solve the problem, and it does not offer a more efficient technique than the ILP approach. The proposed algorithm consists of two components. The first step entails choosing utilization servers for both the database and the application. Next, the second phase entails matching each usage server and its corresponding IoT device . We prove that the proposed algorithm obtains an optimal solution in polynomial time . We compare computation times between the ILP approach and the proposed algorithm. Numerical results show that the proposed algorithm obtains the optimal solution faster than the ILP approach.

A well-known drawback of spectrally-spatially elastic optical networks (SS-EONs) is crosstalk, which increases blocking probability; evaluating blocking probability analytically is challenging due to additional constraints. Not surprisingly, all the existing studies at this time mainly use simulation-based techniques to quantify blocking probabilities in multi-core and multi-mode fibers-based SS-EONs. This paper proposes AnalyticalBP, an exact analytical continuous-time Markov chain model for blocking probabilities considering the crosstalk-avoided approach in SS-EONs. AnalyticalBP generates all the feasible states and their transitions while avoiding inter-core and inter-mode crosstalks for both single and multi-class requests. Single-class requests use the same number of slots, whereas multi-class requests adopt different numbers of slots to satisfy clients’ requirements. When AnalyticalBP is not tractable due to scalability, we introduce an iterative approximate model for a single-hop link. We further extend the single-hop model for a multi-hop network. We compare AnalyticalBP, the iterative approximate model, and Monte Carlo simulation studies for a single-hop link. Numerical results indicate that the performances obtained by AnalyticalBP and the simulation studies, in terms of blocking probabilities and resource utilization, are comparable, and the accuracy of the iterative approximate model is also acceptable.

Given the spring-up of diverse new applications with different requirements in metro optical networks, network slicing provides a virtual end-to-end resource connection with customized service provision. To improve the quality-of-service (QoS) of slices with long-term operation in networks, it is beneficial to reconfigure the slice adaptively, referring to the future traffic state. Considering the busy-hour Internet traffic with daily human mobility, the tidal pattern of traffic flow occurs in metro optical networks, expressing both temporal and spatial features. To achieve high QoS of slices, this paper proposes a node-oriented slice reconfiguration (NoSR) scheme to reduce the penalty of slices, where a gradient-based priority strategy is designed to reduce the penalties of slices overall penalties in reconfiguration. Besides, given that a precise traffic prediction model is essential for efficient slice reconfiguration with future traffic state, this paper presents the model combining the graph convolutional network (GCN) and gated recurrent unit (GRU) to extract the traffic features in space and time dimensions. Simulation results show that the presented GCN-GRU traffic prediction model achieves a high forecasting accuracy, and the proposed NoSR scheme efficiently reduces the penalty of slices to guarantee a high QoS in metro optical networks.

This paper proposes a backup computing and transmission capacity allocation model for virtual networks that minimizes the required backup computing capacity under multiple facility node failures. The proposed model adopts the probabilistic protection, where the probability that the protection fails due to insufficient capacity is restricted not to be greater than a given survivability parameter, by using robust optimization. The conventional model allocates the backup computing capacity by considering the probabilistic protection but allocates the transmission capacity for all backup paths dedicatedly. The proposed model allocates the backup transmission capacity only for the failure patterns that are considered under the probabilistic protection guarantee; the probabilistic protection is considered for both computing and transmission capacity allocation. Reducing the required backup transmission capacity can also reduce the required backup computing capacity, since more feasible solutions for backup computing capacity allocation can exist. We introduce a heuristic algorithm to solve the backup computing and transmission capacity allocation problem. Numerical results show that the proposed model reduces the required backup transmission capacity and enhances the feasibility of allocating virtual networks compared with the conventional model. We also observe that reducing the required backup transmission capacity can lead to reducing the required backup computing capacity.

Middleboxes work as software which runs on a general-purpose server by adopting network function virtualization. The unavailability of middlebox is a key metric. The previous study considers allocating backup servers to middleboxes to reduce the unavailability. While it adopts shared protection to save backup capacity, resource sharing has not been sufficiently explored as each middlebox can only use one backup server. This paper presents a backup allocation model in which a function can be protected by two backup servers and a backup server can protect multiple functions under a shared protection strategy to minimize the maximum unavailability among functions. We use Markov chain to analyze the state transitions and make equilibrium-state equations. By solving them, we obtain the probability of each state of the allocation and compute an unavailability of function. We introduce two algorithms to examine the proposed model; one of them uses the performance bound of the maximum unavailability which is analyzed in this paper. Numerical results show that the proposed model reduces the maximum unavailability by 9.5–51.2% compared to a baseline model that allocates one backup server for each middlebox in our examined cases.

For cloud providers, it is essential to improve the quality of service that depends on the failure probability of protection and the computing capacity cost such as backup resources. Existing studies addressed a protection approach to reduce the required backup capacity by sharing backup capacity among multiple primary resources. Still, the sharing effect is limited, and there is room to enhance it; more primary capacity should share more backup capacity for the enhancement. This paper proposes a model that minimizes the required backup capacity. This model ensures that the backup failure probability, or the probability of unsuccessful backup, of primary resources during multiple simultaneous physical machine (PM) failures does not exceed a given value. In this model, when a PM containing virtual machines (VMs) fails, the VMs in that PM can be recovered by available backup resources in backup PMs. By setting the priority of protecting VMs and backup PMs that each VM is protected by, we obtain the backup failure probability when all available backup resources are used for backup. We introduce heuristic approaches to allocate backup capacity and prioritize the protection to minimize the required capacity while satisfying a given backup failure probability. We introduce a priority policy and a computation policy to reduce the computation time and to be able to deal with larger-size problems. The proposed model can reduce the total required backup capacity compared to the baseline models. We can make the proposed model possible to protect primary resources in larger-size problems.

In network function virtualization, virtual network functions (VNFs) are usually chained in specific orders to generate service function chains (SFCs). Recently, SFC parallelism has been presented to enable VNFs to run in parallel to reduce the end-to-end service delay. Existing works handle the issue of unbalanced parallel branches by assuming predefined linear delay models, which have limitations in efficient resource allocation and deployment cost savings. This paper proposes a deployment model for parallelized SFC that handles the imbalance issue with considering that the delay of each VNF depends on both arriving traffic and allocated computing resources, to improve the flexibility of computing resource allocation. We consider a nonlinear relationship between delay, allocated computing resources, and arriving traffic. We apply VNF sharing to improve the efficiency of resource allocation. We formulate the proposed model as a mixed integer second-order cone programming problem (MISOCP) to minimize the total deployment cost, with satisfying the end-to-end delay requirement. We also introduce a heuristic algorithm to solve the original problem, because the MISOCP approach is intractable to handle larger-size problems in practical time. Numerical results show that the proposed model achieves lower deployment cost than the baseline models.

When mission-critical applications are provided over a network, high availability is required in addition to a low delay network. This paper proposes a multi-homing network design model, named MHND, to balance a low delay and high availability when distributed processing applications use multiple processing servers. MHND maintains the event occurrence order with a multi-homing configuration using conservative synchronization. We formulate MHND as an integer linear programming problem to minimize the delay. We prove that the distributed server allocation problem with MHND is NP-complete. Numerical results indicate that, as a multi-homing number, which is the number of servers to which each user belongs, increases, the availability increases while increasing the delay. Two or more multi-homing can achieve approximately an order of magnitude higher availability compared to that of the conventional single-homing at the expense of a delay increase of 1.25 times. By using MHND, flexible network design is achieved based on the acceptable delay in service and the required availability.

This paper proposes a scheduling model for updating resource allocations for virtualized networks (VNs) without congestion. The proposed model determines the schedule of migrating traffic flows on VNs from old routes to new routes. The model aims to minimize the number of rounds required to complete the update of all existing VNs. We evaluate the performance of model in terms of the percentage of trials where feasible update scheduling exists and the number of rounds required to complete the update. Numerical results show that more rounds are required to achieve congestion-free update when the traffic demand or the number of VNs increases. The number of required rounds tends to remain the same when the maximum number of rounds exceeds a certain value; this observation helps network operators estimate the time required for VN update.

Service chaining provides network services to users by processing packets with a series of virtualized network functions (VNFs). This paper proposes a multi-source multicast service chaining model that guarantees the reliability of services with flexible routing and VNF placement. In order to obtain cost-efficient feasible solutions, we introduce an algorithm that iteratively solves an integer linear programming problem and an algorithm that conducts the VNF placement with a concept of the betweenness centrality. Numerical results show that the proposed model allocates the network and computation resources with the reduction in the total cost compared to the existing model.

This paper proposes a network design approach considering data consistency for a delay-sensitive distributed processing system. The data consistency is determined by collating the own state and the two states of slave servers. If the state is mismatched with other servers, the rollback process is initiated to modify the state to guarantee data consistency. In the proposed approach, the select servers and the master-slave server pairs are determined to minimize the end-to-end delay and the delay for data consistency. We formulate the proposed approach as an integer linear programming problem. We evaluate the delay performance and computation time. The proposed approach reduces the delay for data consistency by 6.8-31.2% compared to that of a typical approach that collates the status of all servers at the master server. The computation time is a few seconds, which is an acceptable time for network design before service launch. These results indicate that the proposed approach is effective for delay-sensitive applications.

In network function virtualization, virtual network functions (VNFs) are usually chained in specific orders to generate service function chains (SFCs). Recently, SFC parallelism has been presented to enable VNFs to run in parallel to reduce the end-to-end service delay. Existing works handle the issue of unbalanced parallel branches by assuming predefined linear delay models, which have limitations in efficient resource allocation and deployment cost savings. This paper proposes a deployment model for parallelized SFC that handles the imbalance issue with considering that the delay of each VNF depends on both the arriving traffic and the allocated computing resources, to improve the flexibility of computing resource allocation. We consider a non-linear relationship between delay, allocated computing resources, and arriving traffic. We apply VNF sharing to improve the efficiency of resource allocation. We formulate the proposed model as a mixed integer second-order cone problem to minimize the total deployment cost, with satisfying the end-to-end delay requirement. Numerical results show that the proposed model achieves lower deployment cost than the baseline models.

This article proposes an optimization model for the network service (NS) mapping and scheduling problem with uncertain processing time in network function virtualization. We model processing time uncertainty through the$\Gamma$-robustness approach, which provides different degrees of robustness against processing time uncertainty. We formulate the problem with the objective to minimize the worst-case makespan over the given uncertainty set. We show the NP-hardness of considered problem. A heuristic that divides the problem into subproblems is presented to tackle it. For the subproblem in which mapping and scheduling decisions are given, we develop an algorithm with polynomial time complexity to calculate the worst-case makespan over the uncertainty set, which has a better scalability than the corresponding mixed integer linear programming (MILP) problem and obtains the same worst-case makespan with the MILP problem. The numerical results show that the proposed model outperforms the conventional model with deterministic parameters in terms of worst-case makespan.

Network function virtualization enables to deploy network services more flexibly with virtual network functions (VNFs). A service provider needs to allocate the required VNFs to hosts, with satisfying different requirements on service delay. It is challenging for service providers to deploy services as efficiently as possible. The previous work has addressed this challenge by studying that multiple services share a VNF instance whose computing capacity is fixed; different VNF instances do not share their computing resources. More efficiency is expected with sharing resources among different VNF instances. This paper proposes a service deployment model for network function virtualization to minimize the deployment cost with capacity sharing and priority queuing. In the proposed model, VNF instances on the same host share the computing capacity in the host by VNF resizing during runtime. In addition, the priority queuing system is applied to each host. We formulate the proposed model as an optimization problem to minimize the service deployment cost. We develop a solution strategy named FlexSize to solve it heuristically in practical time. We evaluate the proposed model with a baseline which does not share the computing capacity among VNF instances in the host. The numerical results show that the proposed model reduces the service deployment cost compared with the baseline.

Network function virtualization enables service providers to flexibly deploy network services. The existing works mainly focus on function placement and capacity allocation without exploring traffic scheduling for service deployment, by adopting a first-in-first-out policy. It introduces inefficiency considering that different services vary in the delay requirements. This paper proposes a service deployment model with priority queuing for both traffic processing and transmission to minimize the deployment cost with satisfying the service delay constraints. We analyze the problem including proving its NP-hardness and the convexity of a subproblem. Based on the analysis, we develop a reinforcement learning-based approach to address the problem in a decomposition manner with a polynomial-time complexity in each episode. Several specific designs are introduced to fit the learning-based approach to the considered deployment problem with priority queueing. The numerical results show that the introduced approach achieves an objective value comparable to the optimal one obtained by brute force search with a computation time$10^{3}$times shorter. Compared to a conventional model with the first-in-first-out policy, the proposed model reduces the deployment cost by adopting a more flexible queueing policy.

This paper proposes a controller placement model that takes into account the load-dependent sojourn time at each controller while considering controller failures in a software-defined network. The sojourn time is expressed by the queuing theory. The sojourn time varies depending on the amount of load arriving at each controller in the proposed model. The proposed model is formulated as a mixed integer second-order cone programming (MISOCP) problem. The controller placement problem studied in this paper is proven to be NP-hard. We develop a heuristic algorithm for the case where the solution to an optimization problem of the proposed model cannot be obtained in practical time. The proposed model is compared with two baseline models presented in the previous research. In the baseline models, the sojourn time does not depend on the amount of load arriving at each controller. Numerical results show that the number of placed controllers becomes smaller in the proposed model than in the baseline models. We also compare results obtained by solving the MISOCP problem to those of the heuristic algorithm. Numerical results show that the heuristic algorithm reduces the computation time required to determine the controller placement, whereas the difference between the number of controllers determined by the heuristic algorithm and the optimal value is at most 4.84%. The number of controllers placed by the heuristic algorithm tends to decrease by considering network centrality.

Service chaining provides network services by flexibly configuring service chains that connect virtualized network functions (VNFs) in the appropriate order so as to satisfy users’ needs. Existing models can be inefficient in terms of consuming network and computation resources since the models do not consider the traffic changes due to VNFs or the models restrict routing and VNF placement. This paper proposes a service chain provisioning model that handles the traffic changes created by VNFs while determining the VNF visit order of each request, request routes, and VNF placement. The service chain provisioning problem is formulated as an integer linear programming (ILP) problem. Three methods for limiting the number of VNF visit order patterns considered in the ILP problem are introduced to shorten the computation time. In order to handle a problem that is intractable with the ILP model, we introduce a greedy algorithm and an algorithm that divides the problem into the VNF placement part and the routing part. Numerical results show that considering the traffic changes due to VNFs yields more efficient consumption of network and computation resources than the alternative of assuming that the traffic amount of each request is constant between the endpoints. The results also show that the computation time can be shortened in our examined scenarios while we obtain the objective value larger by at most 0.4% than the optimal value by limiting the number of visit order patterns considered.

The provisioning cost of a virtualized network (VN) depends on several factors, including the numbers of virtual routers (VRs) and virtual links (VLs), mapping of them on a substrate infrastructure, and routing of data traffic. An existing model, known as the virtual network embedding (VNE) model, determines the embedding of given VN graphs into the substrate infrastructure. When the resource allocation model of the VNE problem is adopted to a single-entity scenario, where a single entity fulfills the roles of both a service provider and an infrastructure provider, an issue of increased costs of VNs and access paths arise. This paper proposes a model for virtualized network graph design and embedding (VNDE) for the single-entity scenario. The VNDE model determines the number of VRs and a VN graph for each request in conjunction with embedding. The VNDE model also determines access paths that connect customer premises and VRs. We formulate the VNDE model as an integer linear programming (ILP) problem. We develop heuristic algorithms for the cases where the ILP problem cannot be solved in practical time. We evaluate the performance of the VNDE model on several networks, including an actual Japanese academic backbone network. Numerical results show that the proposed model designs suitable VN graphs and embeds them according to the volume of traffic demands and access path cost. Compared with the benchmark model, which is based on a classic VNE approach, the proposed model reduces the provisioning cost at most 28.7% in our examined scenarios.

Edge computing is a new paradigm to provide computing capability at the edge servers close to end devices. A significant research challenge in edge computing is finding efficient task offloading to edge and cloud servers considering various task characteristics and limited network and server resources. Several reinforcement learning (RL)-based task-offloading methods have been developed, because RL can immediately output efficient offloading by pre-learning. However, these methods do not take into account clouds or focus only on a single cloud. They also do not take into account the bandwidth and topology of the backbone network. Such shortcomings strongly limit the range of applicable networks and degrade task-offloading performance. Therefore, we formulate a task-offloading problem for multi-cloud and multi-edge networks considering network topology and bandwidth constraints. We also propose a task-offloading method that is based on cooperative multi-agent deep RL (Coop-MADRL). This method introduces a cooperative multi-agent technique through centralized training and decentralized execution, improving task-offloading efficiency. Simulations revealed that the proposed method can minimize network utilization and task latency while minimizing constraint violations in less than one millisecond in various network topologies. It also shows that cooperative learning improves the efficiency of task offloading. We demonstrated that the proposed method has generalization performance for various task types by pre-training with many resource-consuming tasks.

This Special Section features research contributions in the area of robust and reliable networks of the future. Modern network infrastructures must support a growing demand for intensive data processing and high-speed communication, that has led, in the last decade, to a constant evolution towards convergence of networking and computing infrastructures. This convergence was made possible by the introduction of network function virtualization and by the emergence of the Software-Defined Networking (SDN) paradigm, and has enabled new forms of cloud and edge computing to cope with the strict requirements of new services and applications, as those in the realm of the Internet of Things (IoT).

In a cloud, protection through backup of virtual machines contained in physical machines (PMs) reduces damage to users due to failure of PMs, such as hardware malfunctions. However, from a resource cost perspective, it is necessary to reduce the amount of capacity for backup while limiting the probability of unsuccessful protection. Existing studies suggest the method of sharing backup capacity among primary resources, but the amount of capacity reduction required to protect is limited. This paper proposes a backup resource allocation model with two-stage probabilistic protection to minimize the total required backup capacity for multiple simultaneous failures of PMs. In probabilistic protection, backup resources are allocated so that the probability of backup failure does not exceed a given survivability parameter which represents the acceptable probability of backup failure. Probabilistic protection which achieves efficient sharing of backup capacity enables flexible allocation and reduces the required backup capacity. In order to increase the flexibility of backup capacity allocation, the proposed model extends the probabilistic protection to two stages. By dividing the protection into two stages, the weight of probability between stages can be adjusted, enabling more effective capacity sharing. Since it is uncertain which primary PMs fail, we apply robust optimization to the probabilistic protection. By using a table that takes into account the survivability parameter and the failure probability of PMs, the proposed model is formulated as a mixed integer linear programming problem. We prove the NP-hardness of considered problem. A heuristic is introduced to solve the optimization problem. The proposed model can reduce the total required backup capacity compared to the models with dedicated protection and one-stage probabilistic protection. The model can also provide protection in a range of survivability parameters that the model with one-stage probabilistic protection cannot satisfy.

This paper proposes a preventive priority setting model to minimize the worst-case maximum utilization ratio against multiple controller failures in software defined networks. For a set of controllers able to manage a switch, we introduce a priority for each controller to become the main controller that controls the switch. Once the existing main controller fails, the survived controller which has the highest priority works as the main controller. This reassignment is automatically obtained according to the priority setting decided at the network operation start time. In this way, the proposed model provides a prompt recovery with reducing the network instability due to unnecessary controller reassignment. We formulate the proposed model in two different forms, which are an integer linear programming formulation and a min-max formulation. We prove that the considered problem is NP-hard. A basic heuristic is introduced based on the min-max formulation. Its two extensions are further developed considering the accuracy and the computation time in practical computation. Numerical results reveal that, compared to two baselines that sacrifice certain network stability to achieve a more flexible reassignment, the proposed model reduces the network instability by 27% and 52% in average, respectively, with obtaining the comparable maximum utilization ratio.

This paper investigates a service deployment model for network function virtualization which handles per-flow priority to minimize the deployment cost. Service providers need to implement network services each of which consists of one or more virtual network functions (VNFs) with satisfying requirements of service delays. In our previous work, we studied the service deployment model with per-host priority; flows belonging to the same service, for the same VNF, and handled on the same host have the same priority. We formulated the model as an optimization problem, and developed a heuristic algorithm named FlexSize to solve it in practical time. In this paper, we address per-flow priority, in which flows of the same service, VNF, and host have different priorities. In addition, we expand FlexSize to handle per-flow priority. We evaluate per-flow and per-host priorities, and the numerical results show that per-flow priority reduces deployment cost compared with per-host priority.

This paper proposes a virtual network function (VNF) placement model considering both availability and probabilistic protection for service delay to minimize the service deployment cost. Both availability and service delay are key requirements of services; a service provider handles the VNF placement problem with the goal of minimizing the service deployment cost while meeting these and other requirements. The previous works do not consider the delay of each route which the service can take when considering both availability and delay in the VNF placement problem; only the maximum delay was considered. We introduce probabilistic protection for service delay to minimize the service deployment cost with availability. The proposed model considers that the probability that the service delay, which consists of networking delay between hosts and processing delay in each VNF, exceeds its threshold is constrained within a given value; it also considers that the availability is constrained within a given value. We develop a two-stage heuristic algorithm to solve the VNF placement problem; it decides primary VNF placement by solving mixed-integer second-order cone programming in the first stage and backup VNF placement in the second stage. We observe that the proposed model reduces the service deployment cost compared to a baseline by up to 10%, and that it obtains a feasible solution while the baseline does not in some examined situations.

Service chaining concatenates virtual network functions (VNFs) in the network to automatically process packets in an appropriate order. In a situation where network traffic fluctuates, routes of service function chains (SFCs) need to be updated to achieve users' requirements. States of VNF instances that make up the SFC need to be kept consistent when updating the SFC routes. Existing models update SFCs one by one. If these models are adopted to the update of multiple SFCs, the total time required to update all SFCs can be long; problems such as leaving VNFs out-of-date can occur. This paper proposes a model that determines the update schedule of multiple SFCs guaranteeing state consistency. The objective function is minimizing the total time required to update all SFCs. Numerical results show that the total time required to update all SFCs can be shortened by updating multiple SFCs at the same time. The numerical results also show that the total time can be shortened by dividing a large buffer into an adequate number of small buffers.

Sparse mobile crowdsensing is a crowdsensing paradigm that reduces the sensing cost while ensuring data quality by collecting data sparsely and reconstructing desired data using inference algorithms including machine learning algorithms. However, real-time inference of spatial information with sparse mobile crowdsensing has not sufficiently considered the change of temporal characteristics of data. As a result, the accuracy of the reconstructed data can deteriorate over time. Therefore, this paper proposes a framework that periodically updates a machine learning model used for reconstructing data by evaluating the importance of the data in terms of both inference and re-training and giving priority to collecting important data.

This paper proposes an optimistic synchronization algorithm (OSA) based server selection scheme with successive participation scenario. In the scenario, we introduce a participating-domain segmentation and determe recommended servers before user participation. Numerical results indicate that the proposed scheme reduces the latency compared to the non-domain segmentation approach (conventional scheme) and overcomes latency fluctuation.

This paper proposes a gradual program file placement control method using intermediate solutions output by an optimization solver. The computation resources in the platform can be efficiently utilized by reconfiguring the program file placement following the intermediate solutions. We present two policies, which differ on when to reconfigure the platform using intermediate solutions. Simulation results show that the proposed method achieves less cumulative number of program files placed in the platform, in exchange for the increase in the reconfiguration cost.

This paper proposes a robust function deployment model against uncertain recovery time with satisfying an expected recovery time guarantee in a cost-efficient manner. We consider that each node fails with a workload-dependent failure probability, which is a non-decreasing function that reveals the empirical relationship between the workload and the failure probability. The preventively deployed backup resources can recover an unavailable function hosted by a failed node in a period of time, which is related to the backup strategies and failure and recovery scenarios. We introduce an uncertainty set that considers the upper and lower bounds of the recovery time of a function by each node that protects it and the upper bound of the average recovery time among nodes. The robust optimization technique is applied to handle the worst case of expected recovery time satisfying a time guarantee under an uncertain recovery time. With this technique, the model is formulated as a mixed integer linear programming problem. The numerical results reveal that the proposed model saves the deployment cost on average 24% compared to a baseline that uses the deterministic recovery time in our tested cases.

This paper proposes an availability-aware service provisioning model with backup sub-chain-enabled sharing in network function virtualization to minimize the deployment cost. A sub-chain consists of a set of ordered VNFs that corresponds to a part of or the whole function chain of a service. Different from a conventional model in which a backup sub-chain is dedicated to protecting a primary sub-chain of one service, the proposed model allows the backup sub-chain sharing among services to reduce the deployment cost. Due to the complexity of the investigated problem, a heuristic is designed to tackle it. The numerical results show that the proposed model achieves lower deployment cost with satisfying the availability requirement than the conventional one.

Optical regenerators are beneficial in resource utilization as they provide additional functionalities, such as modulation format (MF) and spectrum conversion, besides signal regeneration. In spectrally-spatially elastic optical networks (SS-EONs), regenerators can perform core and mode switching, which further improves the spectrum utilization. For the first time, this paper proposes a regenerator-aware routing, spectrum, core, and mode allocation model while avoiding inter-mode and inter-core crosstalks during resource allocation to enhance the spectrum utilization in SS-EONs. The proposed model performs core/mode switching operations at the regeneration sites along with spectrum and MF conversions. Apart from the regeneration sites, the proposed model maintains the spectrum continuity, spectrum contiguity, core continuity, and mode continuity constraints in the remaining intermediate nodes. We model the regenerator-aware resource allocation as an integer linear programming to minimize the highest utilized spectrum slot index under the condition that a limited number of regenerators with their placement are given in the network. We introduce a heuristic when the optimization problem is not tractable. Numerical results indicate that the proposed model improves resource utilization compared to a benchmark model that does not consider core and mode switching.

To accommodate the exponential growth of the inter-network services, infrastructure as a service (IaaS) allows the different parties to share the physical infrastructure of the optical network resources using network virtualization. The virtual optical network embedding (VONE) enables efficient resource virtualization to map several virtual optical network (VON) requests over a substrate optical network. On the other hand, the spectrally-spatially elastic optical network (SS-EONs) is becoming a promising solution for the increasing volume of the demanded traffic due to its higher fiber capacity. For the first time, this paper proposes a routing, spectrum, core, and mode allocation (RSCMA) model for VONE over SS-EONs while avoiding inter-core and inter-mode crosstalks. The proposed model allocates spectrum for VON requests over the substrate SS-EONs while maintaining the spectrum continuity, spectrum contiguity, core continuity, and mode continuity constraints. We model the RSCMA model for the VONE problem over SS-EONs as an integer linear programming (ILP) to maximize the number of VON requests placed in the substrate network. We introduce a heuristic when the optimization problem is not tractable for large networks. Numerical results indicate that the performances of the ILP and heuristic approaches are comparable and the computation time of the heuristic approach is much smaller than that of the ILP approach.

Nowadays, quantum-key-distribution (QKD) gains popularity for providing high-security in high-capacity spectrally-spatially elastic optical network (SS-EONs). Considering QKD in SS-EONs requires addition resources for classical and quantum channels and introduces addition noises. The key challenge during resource allocation in QKD-enabled SS-EONs is to deal with the background noises generated by classical and quantum channels along with inter-core and inter-mode XT while reducing the highest utilized spectrum slot index. For the first time, this paper proposes a routing, spectrum, core, and mode allocation (RSCMA) model in QKD-enabled SS-EONs while avoiding intercore and inter-mode XT and the noises generated by the classical and quantum channels. The proposed model allocates spectrum for the data and classical channels for each request maintaining the spectrum continuity and contiguity constraints. It selects a dedicated mode for the quantum channels along the routing path for each request. We model the QKD-enabled resource allocation as an integer linear programming (ILP) to minimize the highest utilized spectrum slot index. We introduce a heuristic when the optimization problem is not tractable for large networks. Numerical results indicate that the values of the highest utilized spectrum slot index using the ILP and heuristic approaches are comparable and the computation time of the heuristic approach is much smaller than that of the ILP approach.

Queuing delay is a dynamic network parameter that plays an important role in defining the performance of Internet applications over an end-to-end path. However, measurement of queuing delay is challenging because it requires a large infrastructural support from the path under test. In this paper, we propose an active scheme to measure queuing delay on a router using a probe-gap model. The scheme uses a popular data-clustering algorithm to process its data samples; therefore, its measurement efficacy is not dependent on the issues related to infrastructural access, certain variations (e.g., compression) in the probe gaps, and the number of clusters in the data processing. Here, we present a detailed evaluation of the scheme against the current state-of-the-art on a single-hop path through ns-3 simulation. Our results show that the proposed scheme is robust, consistent, quick, and highly accurate under different traffic conditions.

Network operators can operate services in a flexible way with virtual network functions thanks to the network function virtualization technology. Flow partition allows aggregated traffic to be split into multiple parts, which increases the flexibility. This paper proposes a service deployment model with flow partition to minimize the total deployment cost with meeting service time delay requirements. A virtual network function of a service is allowed to have several instances, each of which hosts a part of flows and can be shared among different services, to reduce the initial and proportional cost. We provide the mathematical formulation for the proposed model. A heuristic algorithm is introduced to solve the original problem in practical time by decomposing it into several steps; each step handles a convex problem. The numerical results reveal that the proposed model saves the total deployment cost compared to the conventional one. It improves the maximum admissible traffic scale by 23% in average in our examined cases.

Multi-access edge computing (MEC) enables users to exploit the resources of cloud computing at a base station (BS) in proximity to the users where an MEC server is hosted. While we have advantage of being able to communicate with low latency and small network load in MEC networks, the resources in BSes are limited. One challenge is where to provide users with services from to make efficient use of resources. Furthermore, to enhance the reliability of MEC system, the case that a BS fails needs to be considered. This paper proposes a service placement and user assignment model with preventive start-time optimization against a single BS failure in MEC networks. The proposed model preventively determines the service placement and user assignment in each BS failure pattern to minimize the worst-case penalty which is the largest penalty among all failure patterns. We formulate the proposed model as an integer linear programming problem. We introduce two algorithms, one is the greedy algorithm with allocation upgrade and the other is with allocation upgrade and preemption, to solve the problem. The results show that the introduced algorithms obtain a solution with the smaller worst-case penalty than the benchmark in a practical time.

This paper proposes a controller placement model that takes into account the load-dependent sojourn time at each controller while considering controller failures. The sojourn time is expressed by the queuing theory. The sojourn time varies depending on the amount of load arriving at each controller in the proposed model. The proposed model is formulated as a mixed integer second-order cone programming problem. The proposed model is compared with two baseline models presented in the previous research. In the baseline models, the sojourn time does not depend on the amount of load arriving at each controller. Numerical results show that the number of placed controllers becomes smaller in the proposed model than in the baseline models. This indicates that, since the sojourn time in the proposed model varies according to the amount of load at a controller, the effect of the load-dependent sojourn time at a controller tends not to become dominant over that of the propagation delay, which enables a switch to connect to more distant controller than that in the baseline models.

Middleboxes work as software which runs on a general-purpose server by adopting a network function virtualization. The unavailability of middlebox is a key metric. The previous study considers allocating backup servers to middleboxes to reduce the unavailability. While it adopts shared protection to save backup capacity, resource sharing has not been sufficiently explored as each middlebox can only use one backup server. This paper presents a backup allocation model in which a function can be protected by two backup servers and a backup server can protect multiple functions under a shared protection strategy to minimize the maximum unavailability among functions. We use Markov chains to analyze the state transitions and make equilibrium-state equations. By solving them, we obtain the probability of each state of the allocation and compute an unavailability of function. We introduce two algorithms to examine the proposed model; one of them uses the performance bound of the maximum unavailability which is analyzed in this paper. Numerical results show that the proposed model reduces the maximum unavailability by 13.9-51.2% compared to a baseline model that allocates one backup server for each middlebox in our examined cases.

This paper proposes a backup resource allocation model with two-stage probabilistic protection to minimize the total required backup capacity for multiple simultaneous failures of physical machines (PMs). Probabilistic protection ensures that the probability that the PM used for backup fails to backup due to lack of computing capacity does not exceed a given survivability parameter. The proposed model protects the primary virtual machines by allocating computing capacity to backup PMs with probabilistic protection. Since it is uncertain which primary PMs fail, we apply robust optimization to the probabilistic protection. By using a table that takes into account the survivability parameter and the failure probability of PMs, this proposed model is formulated as a mixed integer linear programming problem. The proposed model extends the probabilistic protection to two stages; the VMs that fail to be protected in the first stage are protected in the second stage to achieve the probabilistic protection with the final survivability parameter. This model can reduce the total required backup capacity compared to the conventional model with one-stage probabilistic protection.

This paper proposes an optimization model to derive a resilient virtual network function (VNF) allocation aiming to maximize the number of accepted requests with considering VNF diversity and ensuring the requirements of node fault tolerances in a dynamic scenario, where the requests have random requirements, arriving and releasing time. The model considers fault tolerance assurance and satisfies the service requirements under different error patterns. The allocation provided by the proposed model ensures the required amount of processing ability in the situation that there are several failed nodes. The node fault tolerance can be set variously for different requirements of services. The proposed model selects and instantiates suitable replicas from the pools of replicas, and then determines the locations of these replicas instances. We develop a reinforcement learning approach for solving the proposed model including the design of the learning environment and the reward shaping. The numerical results show that the proposed model increases the number of accepted requests with ensuring the resiliency of the functions compared with baseline models in the examined cases, where the allocation of a request can be determined in tens of milliseconds.

Prompt function deployment and management is a key role in network function virtualization to improve the continuity and reliability of network services. Kubernetes is a system to deploy and manage functions automatically. Existing tools in Kubernetes do not provide automatic function deployment and management in a real-time and optimal manner. It does not provide a resource type to manage the migratable resource, either. This paper designs and implements a two-layer controller structure in Kubernetes to achieve the function deployment in a limited computation time with considering resource migration for allocation optimality. A controller in the lower layer manages the Pods for an intermediate allocation with a model or a heuristic algorithm to respond to requests promptly. A controller in the upper layer manages instances by optimizing resource allocations with considering resource migration; it maintains the Pods by keeping the current state (intermediate allocation) consistent with the desired state (optimal allocation). Our demonstration validates that the controller automatically manages the resources promptly and correctly.

This article proposes a primary and backup resource allocation model that provides a probabilistic protection guarantee for virtual machines against multiple failures of physical machines in a cloud provider to minimize the required total capacity. A physical machine allocates both primary and backup computing resources for virtual machines. When any failure occurs, the survived physical machines with preplanned backup resources recover the virtual machines on the failed physical machines and take over the workloads. The probability that the protection provided by a physical machine does not succeed is guaranteed within a given number. Providing the probabilistic protection can reduce the required backup capacity by allowing backup resource sharing, but it leads to a nonlinear programing problem in a general-capacity case against multiple failures. We apply robust optimization with extensive mathematical operations to formulate the primary and backup resource allocation problem as a mixed integer linear programming problem, where capacity fragmentation is suppressed. We prove the NP-hardness of considered problem. A heuristic is introduced to solve the optimization problem. The results reveal that the proposed model saves about one-third of the total capacity in our examined cases; it outperforms the conventional models in terms of both blocking probability and resource utilization.

To support the fifth-generation bandwidth-hungry applications, such as the Internet of Things, virtual reality, augmented reality, and cloud computing, elastic optical networks have become the most promising infrastructure that allocates bandwidths for services flexibility. Fragmentation caused by dynamic resource allocation deteriorates the availability of resources in networks, increasing the blocking of requests. The fragmentation occurs not only in the used path but also in the neighboring links that are not included in the used path; they are connected to the used path. This paper proposes a service-driven fragmentation-aware (SDFA) resource allocation scheme to enhance resource utilization by avoiding fragmentation with the joint consideration of the used path and neighboring links. A service-driven fragmentation metric (SDFM) is, for the first time, presented to estimate the fragmentation in the used path and neighboring links. The SDFA scheme prefers to assign services at the spectrum slots, which leads to the minimum value of SDFM. Simulation results indicate that SDFA outperforms four conventional fragmentation-aware resource allocation schemes in terms of blocking probability and resource utilization due to a lower fragmentation in the network.

Signal transmission using counter-propagation nowadays is adopted to enhance resource utilization in spectrally-spatially elastic optical networks (SS-EONs), where inter-mode and inter-core crosstalks always degrade signal quality and become bottlenecks for high transport capacity. In this paper, we propose BPRIA for the first time, a Bi-Partitioning-based crosstalk-avoided Resource Identification and Allocation scheme in SS-EONs to enhance resource utilization while suppressing both inter-mode and inter-core crosstalks. We introduce a bi-partitioning optimization problem with vertex elimination to maximize the number of non-adjacent cores and modes in each partition of the bipartite graph; two distinct sets of cores and modes are used in a counter-propagation manner for lightpath allocation. The optimization problem is formulated as an integer linear programming (ILP) problem, and we prove that the optimization problem is NP-complete. An algorithm for core-mode-spectrum allocation is developed considering two distinct sets of cores and modes obtained by ILP to serve lightpath requests while avoiding inter-mode and inter-core crosstalks. For dynamic scenarios, we present core-mode-spectrum allocation. Numerical results reveal that the blocking probability is reduced by BPRIA in SS-EONs, and it enhances traffic admissibility in the network.

This paper proposes a survivable virtual network embedding model over elastic optical networks considering shared protection against any single substrate node or link failure. A virtual network request is embedded in the substrate network with allocating the primary and backup resources which are node-disjoint. Modulation selection and spectrum allocation with constraints from elastic optical networks are considered in the embedding procedure. We consider the backup computing and transmission resource sharing to reduce the required backup resources. We formulate the proposed model as an integer linear programming problem to minimize the rejection ratio for a given set of virtual network requests. We introduce a greedy-based approach that promotes resource sharing to handle the larger-size problem in a practical time. In order to further improve the performance on the objective value, a deep reinforcement learning-based approach with polynomial time complexity in each episode is developed to solve the problem in multiple stages. We design a dedicated learning agent for each stage considering the problem property. We analyze the usages of different approaches based on performance evaluation. The numerical results show that, compared to a conventional model that adopts dedicated protection, the proposed model with shared protection reduces the rejection ratio by 15% on average in our examined cases.

Packet processing tasks in network functions require high-performance memory systems to understand the packet information, update the packet content, and search the databases. While network virtualization is expected to bring flexible and adaptive network with reduced cost by using commercial off-the-shelf (COTS) hardware and programmable data plane technology, network function performance suffers from the poor memory systems in COTS computers and lack of scalability in programmable hardware devices. This paper proposes a memory network architecture for packet processing based on memory-centric, disaggregated computing. Unlike processor-centric architecture in today’s COTS computers, the memory network consists of multiple memory devices, where processing for the incoming packets is completed. The proposed architecture reduces packet processing latency by eliminating communication between the processor devices and the memory devices. Also, the proposed architecture provides scalability of hardware resources by dynamic memory device allocation depending on the complexity of the network function, memory-intensiveness of packet processing, and traffic load. The numerical results show that the proposed architecture reduces accumulated latency for memory accesses and increases throughput compared to the conventional, processor-centric architectures, where every memory access requires communication between the processor devices and the memory devices. The proposed architecture also reduces latency and increases throughput by allocating additional memory devices to memory-intensive tasks.

Recently, space-division multiplexing (SDM) has been incorporated with elastic optical networks (EONs) to enhance the fiber capability, which forms space-division multiplexing-based elastic optical networks (SDM-EONs). During transmission of optical signals through multi-core fibers, inter-core crosstalk (XT) and intra-core physical layer impairments (PLIs) arise, which deteriorate the signal quality. Existing models typically handle inter-core XT and intra-core PLIs separately and set a single XT threshold for each modulation format, which results in an unacceptable lightpath due to the degradation of signal quality or leads to inefficient spectrum utilization. This paper proposes a routing, modulation, spectrum, and core allocation (RMSCA) model for SDM-EONs to consider inter-core XT and intra-core PLIs jointly. For each modulation format, it sets different XT thresholds and transmission reaches according to inter-core XT and intra-core PLIs. An optimization problem is formulated as an integer linear programming (ILP) problem. We prove that the RMSCA decision problem is NP-complete. We introduce a heuristic algorithm when the ILP problem is not tractable. Numerical results demonstrate that, by setting several XT limits for each modulation format, the proposed model increases spectrum efficiency compared to a benchmark model based on the literature.

This Special Section (SI) features the latest research contributions regarding recent advances in the design and management of reliable communication networks. Communication networks are constantly increasing their complexity and scale to satisfy the requirements of network services. The current trend of convergence of networking and computing infrastructures (as in today’s cloud systems and softwarized networks) calls for novel advanced strategies and solutions to support reliable services, as the development of new data-driven solutions for reliable network automation and self-diagnostic tools to ensure resilient network management.

This paper proposes a sub-chain-enabled coordinated protection model for the availability-guaranteed service function chain (SFC) provisioning, which considers the availability of each component to constitute an SFC, including links and VNFs. Unlike conventional protection models providing certain protection for the whole chain, the proposed model configures sub-chains for each SFC and provides proper protection for each sub-chain to achieve the required availability in a cost efficient way. We formulate the proposed model as an optimization problem to minimize the deployment cost. A game approach is presented to tackle the problem. The numerical results show that the proposed model outperforms the conventional ones in terms of deployment cost; the game approach has scalability of tackling the proposed model as the problem size increases.

This paper proposes an optimization model to derive a primary and backup resource allocation considering a workload-dependent failure probability to minimize the maximum expected unavailable time (MEUT). The workload-dependent failure probability is a non-decreasing function which reveals the relationship between the workload and the failure probability. The proposed model adopts hot backup and cold backup strategies to provide protection. The cold backup strategy is a protection strategy, in which the requested loads of backup resources are not activated before failures occur to reduce resource utilization with the cost of longer recovery time. The hot backup strategy is a protection strategy, in which the backup resources are activated and synchronized with the primary resources to recover promptly with the cost of higher workload. We formulate the optimization problem as a mixed integer linear programming (MILP) problem. We prove that MEUT of the proposed model is equal to the smaller value between the two MEUTs obtained by applying only hot backup and cold backup strategies with the same total requested load. A heuristic algorithm inspired by the water-filling algorithm is developed with the proved theorem. The numerical results show that the proposed model suppresses MEUT compared with the conventional model which does not consider the workload-dependent failure probability. The developed heuristic algorithm is approximately 105times faster than the MILP approach with 10−2performance penalty on MEUT.

Fault tolerance and load balancing are two key roles in resource allocation against failures. This paper proposes a primary and backup resource allocation model with preventive recovery priority setting to minimize a weighted value of unavailable probability (W-UP) against multiple failures. W-UP considers the probability of unsuccessful recovery and the maximum unavailable probability after recovery among physical nodes. We consider that each node fails with a workload-dependent failure probability; each failure pattern occurs with a probability. The workload-dependent failure probability is a non-decreasing function revealing an empirical relationship between the workload and the failure probability for each physical node. We introduce a recovery strategy to handle the workload variation which is determined at the operation start time and can be applied for each failure pattern. Once a failure pattern occurs, the recoveries are operated according to the priority setting to promptly recover the functions hosted by failed nodes. We also discuss an approach to obtain unsuccessful recovery probability with considering the maximum number of arbitrary recoverable functions by a set of available nodes without the priority setting. We formulate the optimization problem as a mixed integer linear programming (MILP) problem. We develop a heuristic algorithm to solve larger size problems in a practical time. The developed heuristic algorithm is approximately 729 times faster than the MILP approach with 1.6% performance penalty on W-UP. The numerical results observe that the proposed model reduces W-UP compared with baselines.

In distributed processing for communication services, a proper server selection scheme is required to suppress delay by ensuring the event occurrence order. Although a conservative synchronization algorithm (CSA) has been used in this issue, an optimistic synchronization algorithm (OSA) can be a potential candidate for synchronizing distributed systems. In comparison with CSA, which reproduces events in occurrence order before processing application, OSA can be feasible to realize low delay communication as the processing events arrive sequentially. This paper proposes an optimal server selection scheme considering OSA for distributed processing systems to minimize end-to-end delay under the condition that the holding time for application status is limited. In other words, the end-to-end delay is minimized based on the allowed rollback time for application design or quality-of-service. Numerical results indicate that the delay of the proposed scheme can be reduced by up to a quarter compared to that of the conventional scheme that is based on CSA.

This paper proposes a service deployment model for network function virtualization to minimize the deployment cost with capacity sharing and priority queuing. A service provider needs to allocate the the required virtual network functions (VNFs) to hosts, with satisfying different requirements on service delay. In the previous work, it has been studied that multiple services share a VNF instance whose computing capacity is fixed; different VNF instances do not share their computing resources. In the proposed model, VNF instances on the same host share the computing capacity in the host by VNF resizing during runtime. In addition, the priority queuing system is applied to each host. We formulate the proposed model as an optimization problem to minimize the service deployment cost. We develop a solution strategy named FlexSize to solve it heuristically in practical time. We evaluate the proposed model with a baseline which does not share the computing capacity among VNF instances in the host. The numerical results show that the proposed model reduces the service deployment cost compared with the baseline.

In the modern era, we need to deploy several functionalities either on central cloud servers or edge cloud servers to provide Internet of Things (IoT)-based services via a wide-area network. Typically a huge database and several non-real-time functions are deployed on the central cloud servers. On the other hand, real-time functions are deployed on edge cloud servers. For delay-sensitive services, this approach has an issue in completing the analysis with the latest information when the delay between the central and edge cloud servers is large. In this paper, we propose an allocation scheme of database and applications, which minimizes the delay from the latest update of the database to analyze the real-time data from IoT devices. In the proposed scheme, the delay is minimized considering two constraints of each server, which are maximum accommodating capacity of the IoT devices and whether the database function can be deployed. Numerical results observe that the proposed scheme reduces the delay of analysis compared to the conventional scheme. These results indicate that the proposed scheme can configure a low-delay network for delay-sensitive IoT services with data analysis.

Visual surveillance of dynamic objects on roads has been developed to ensure road safety for people. Particularly, vehicle tracking is considered as a key technology for the road safety; studies on multi-object tracking (MOT) are being actively pursued. However, when MOT is performed, raw vision data are not always available because of the technical limitation or the privacy concern of the system; MOT needs to be performed only using the coordinates obtained from the object detector without using features extracted from raw image data such as color of vehicles, which degrades the accuracy of MOT to the unsatisfactory level for road safety. This paper proposes an MOT scheme for moving vehicles that is inspired by cell tracking using the Viterbi algorithm. The proposed scheme extends the Brownian motion model, which was used in the base scheme of cell tracking, by weighting probability transitions in accordance with the direction of travel of vehicles on the road. We evaluate the proposed scheme using simulated vehicle-traffic data and verify that the proposed scheme performs better than benchmark schemes in terms of the accuracy of MOT. We also demonstrate an example of how the proposed scheme works well for real vehicle-traffic data.

This paper proposes a backup resource allocation model for virtual network functions (VNFs) to minimize the total required backup computing capacity with considering the service delay. If random failures occur to primary hosts, the VNFs in failed hosts are recovered by backup hosts, where the allocation is determined in advance. We introduce the probabilistic protection, where the probability that the protection provided by a backup host fails is limited within a given value; it allows backup resource sharing to reduce the total required computing capacity. The previous work formulated the backup resource allocation problem without considering the service delay as a mixed integer linear programming (MILP) problem by adopting the robust optimization. We consider the delay of services, which consists of networking delay between hosts and processing delay in each requested VNF. The probability that the total delay of a service exceeds its threshold is constrained within a given value. To solve the problem with the delay constraint, we introduce an algorithm with two methods to make the MILP problem be aware of the service delay. The results observe that, compared to the baseline, the proposed model can reduce the total required backup capacity of computing resource.

This paper proposes a service chain provisioning model considering traffic changing effects of virtualized network functions (VNFs) while determining the VNF visit order of each request, routes of requests, and VNF placement. The service chain provisioning problem is formulated as an integer linear programming (ILP) problem. Three methods of limiting the number of VNF visit order patterns considered in the ILP problem are introduced in order to shorten the computation time. These methods select visit order patterns so as to suppress the sum of the traffic amount reserved by each request on its route. Numerical results show that the consumption of network and computation resources becomes more efficient by considering traffic amount changed by VNFs than the case assuming the traffic amount of each request to be constant end-to-end. The results also show that the computation time can be shortened in our examined scenarios while we obtain the objective value larger by at most 0.4% than the optimal value by limiting the number of visit order patterns.

This paper proposes an optimization model to derive the resilient virtual network function allocation in service function chains aiming to reduce the recovery time during the migrations from the primary functions to backup functions. We consider k-fault-tolerance assurance and satisfy the service requirements under different error patterns in this model. The allocation provided by the proposed model ensures that the processing ability satisfies the requirements even though there are k failed nodes in the network. Diversity splits a single VNF into a pool of replicas with different specifications. The diversity of both primary and backup functions are considered. Redundancy is used for recovering the failed functions. We formulate the proposed model as a mixed integer linear programming problem to select suitable replicas from the pools of replicas and decide the locations of these replicas for both primary and backup functions. The objective of the proposed model is to minimize the sum of the maximum recovery time among functions under all possible failure patterns which have k node failures. The numerical results show that the proposed model reduces the recovery times of VNFs with ensuring the resiliency of the functions compared with baseline models in the examined cases. We give some methods to improve the maximum resiliency at last.

Space-division multiplexing-based elastic optical networks (SDM-EONs) enhance the fiber capacity. Inter-core crosstalk (XT) and intra-core physical layer impairments (PLIs) occur in the multi-core fiber and degrade the optical signal. An existing model separately considers inter-core XT and intra-core PLIs, and sets a single XT threshold to each modulation format. This can lead to an unacceptable lightpath due to signal degradation or the occurrence of spectrum inefficiency. This paper proposes a routing, modulation, spectrum, and core allocation (RMSCA) model, which jointly considers inter-core XT and intra-core PLIs for SDM-EONs. The proposed model sets multiple XT thresholds for each modulation format based on inter-core XT and intra-core PLIs. We present an optimization problem and formulate it as an integer linear programming (ILP) problem. We introduce a heuristic algorithm for a network where the ILP problem is not tractable. Numerical results observe that the proposed model improves the spectrum efficiency by setting multiple XT thresholds to each modulation format.

Network function virtualization enables service providers to flexibly provision services with virtual network functions. Traffic uncertainty typically exists in a network, which can degrade the performance of a virtual network function. This paper proposes a robust virtual network function deployment model against the traffic uncertainty to minimize the total deployment cost with satisfying the service delay constraint. A virtual network function instance is allowed to be shared by different services to reduce the initial and proportional costs. We describe the traffic uncertainty from different aspects with considering the characteristics in the context of network function virtualization. We formulate the robust deployment problem as a mixed integer second-order cone programming problem. A heuristic algorithm is introduced to solve the problem polynomially by decomposing the original problem to several convex problems. The numerical results reveal that the proposed model saves the deployment cost in average 27% compared to a baseline that uses the deterministic traffic arrival rate, in our examined scenarios.

Packet processing tasks in network functions issue memory requests to understand the packet information, update the packet content, and search the databases, which requires high-performance memory systems. While network functions virtualization (NFV) is expected to reduce the cost of network infrastructure by replacing dedicated network equipment with commercial off-the-shelf (COTS) hardware and virtual network functions (VNFs), VNF performance suffers from the poor memory systems that lack function-dedicated memories and memory parallelism in COTS servers. While several works presented parallel memories for packet processing based on 3 dimensional (3D)-stacked dynamic random access memories (DRAMs), data transfer latency between the processors and memories are not considered. Although there are processing-in-memory (PIM) architectures that offload a part of processing in memories to reduce data transfers, the majority of processing is still in the processors, which requires data transfers for multiple packet processing tasks. This paper proposes a memory network architecture using 3D-stacked DRAMs to increase throughput and reduce accumulated latency when there are multiple packet processing tasks. Packets that enter the memory network receive packet processing at each 3D-stacked DRAM without data transfers between the processors and memories. The evaluation results show that the proposed architecture increases throughput and reduces accumulated latency for memory accesses and data transfers when there are multiple packet processing tasks, compared to the conventional architecture with 3D-stacked DRAM-based parallel memory, where every memory access requires data transfers.

Resource allocation and management is a key role in network function virtualization to improve the reliability of network services. Kubernetes is a system to deploy and manage the virtual network functions automatically. Existing tools in Kubernetes does not provide a resource type to define the backup Pods. It does not provide automatic resource management based on the user requests for the backup Pods, either. This paper designs and implements a custom resource and the corresponding controller in Kubernetes to manage the primary and backup resources of network functions. The custom resource is a set of Pods with different types, which includes primary, hot backup, and cold backup Pods. The controller manages the set of Pods and maintains the current state of the different types of Pods to keep the current state consistent with the desired state of each type of Pod. Demonstration validates that the controller automatically manage the primary and backups resources correctly.

Diversity in network function virtualization is to use a group of thin replicas to provide the network services under the required processing ability. Redundancy is to provide a certain number of replicas against function failures and improve network reliability. Kubernetes is a system to deploy and manage virtual network functions automatically. Existing tools in Kubernetes do not provide a resource type to provide required functions jointly considering VNF diversity and redundancy. This paper designs and implements a custom resource and the corresponding controller in Kubernetes to manage the VNF diversity and redundancy jointly. The controller selects suitable replicas from a pool of replica templates to satisfy the required processing ability with the minimum required number of replicas and converts the backup functions to the primary functions when the primary functions cannot provide the required ability. Demonstration validates that the controller automatically manages the resources correctly, improves the resource utilization, and increases the number of acceptable requests.

This paper proposes a two-level processing scheme for three-dimension-image sensing. The first level processing selects only spatial regions needed for a smart monitoring task to reduce the total volume of data traffic. The second level processing integrates multiple (physical) image sensors into a virtual one to improve the delay and jitter performance in the realtime transmission of data from sensors to the cloud server. We develop a prototype system to implement the proposed scheme. Our demonstration validates that the proposed processing scheme works better than the benchmarks which do not adopt the two-level processing.

Fragmentation with crosstalks is the major obstacle in spectrally-spatially elastic optical networks, which suppresses resource utilization while degrading the quality-of-transmission. To overcome this issue, this article proposes, for the first time, a proactive fragmentation management scheme based on batch processing while satisfying both inter-core and inter-mode crosstalks to enhance resource utilization. The proposed scheme adopts a batch processing method to create batches of lightpath requests received within a time threshold to utilize spectrum resources effectively. In batch processing, lightpath requests are prioritized based on the number of links in their routes and required slots. To maintain fairness in batch processing, when any request is rejected, the proposed scheme triggers a procedure that gives an equal opportunity to all arriving requests within the threshold, irrespective of numbers of hops and requested capacities, for allocation. We formulate the static batch processing of lightpath requests (SBPLR) as an integer linear programming (ILP) problem. We prove that SBPLR is an NP-Complete problem. We introduce a heuristic solution when ILP is intractable. To serve lightpath requests in each batch while avoiding inter-core and inter-mode crosstalks, we develop a core-mode-spectrum allocation algorithm. We present a dynamic batch processing based fragmentation management approach. Numerical results indicate that the proposed scheme outperforms the benchmark schemes.

Network function virtualization provides an efficient and flexible way to implement network functions deployed in middleboxes as software running on commodity servers. However, it brings challenges for network management, one of which is how to manage the unavailability of middleboxes. This article proposes an unavailability-aware backup allocation model with the shared protection to minimize the maximum unavailability among functions. The shared protection allows multiple functions to share the backup resources, which leads to a complicated recovery mechanism and makes unavailability estimation difficult. We develop an analytical approach based on the queueing theory to compute the middlebox unavailability for a given backup allocation. The heterogeneous failure, repair, recovery, and waiting procedures of functions and backup servers, which lead to several different states for each function and for the whole system, are considered in the queueing approach. We analyze the performance bounds for a given solution and for the optimal objective value. Based on the developed analytical approach and the performance bounds, we introduce two heuristics to solve the backup allocation problem. The results reveal that, compared to a baseline model, the proposed unavailability-aware model reduces the maximum unavailability 16% in average in our examined scenarios.

This paper proposes a backup computing and transmission resource allocation model for virtual networks with providing a probabilistic protection against multiple facility node failures. The proposed model aims to find the allocation to minimize the required backup computing capacity, which guarantees the probability that the protection fails due to insufficient reserved backup computing capacity within a given value. The previous study only considers the backup computing resource allocation for virtual nodes regardless of the network aspects. In this paper, backup transmission resource allocation is incorporated, where the required backup transmission capacity can affect the required backup computing capacity. We analyze backup transmission resource sharing in the case of multiple facility node failures to compute the minimum required backup transmission capacity. A heuristic algorithm is introduced to solve the problem; especially, several techniques based on graph theory are developed to handle the problem with full backup transmission resource sharing. The result observes that the proposed model outperforms a baseline with dedicated protection for computing resource. Furthermore, the application scenarios for the proposed model with different degrees of backup transmission resource sharing are analyzed. With our analyses, a network operator can set an appropriate degree of backup transmission resource sharing based on practical requirements.

This paper proposes a main and secondary controller assignment model against multiple controller failures in software defined networks considering latency between switches and controllers. The survivability guarantee of each switch is satisfied by assigning a set of controllers, where one of them works as the main controller to control the switch. Given assigned controllers, we introduce a policy-based approach to automatically specify the main controllers in each failure pattern, which leads to a lightweight configuration on a switch. We define the average-case expected latency, the worst-case expected latency, and the expected number of switches within a latency bound, as three objectives to be optimized in three different problems. We prove that a low latency first policy achieves the optimal objective for each considered problem. We formulate the proposed controller assignment model with different goals as three mixed integer linear programming problems. We prove the NP-completeness for all the three problems. A greedy algorithm with polynomial time complexity is developed; we show that it provides a 1/2-approximation for the case without the survivability guarantee constraint. The numerical results observe that the proposed model obtains the optimal objective value with the computation time about$10^{2}$times shorter than that of a baseline that introduces decision variables to determine the main controllers.

This article proposes a network design model to minimize the worst-case network congestion against multiple link failures, where open shortest path first link weights are determined at the beginning of network operation. In the proposed model, which is called the preventive start-time optimization model against multiple link failures (PSO-M), the number of multiple link failure patterns to support is restricted by introducing a probabilistic constraint calledprobabilistic guarantee. If the total probability of non-connected failure patterns does not exceed a specified probability, PSO-M provides a feasible solution of link weights. Otherwise, no feasible solution can be obtained. We introduce an extended model of PSO-M, called PSO-M with link reinforcement (PSO-MLR), where links are reinforced under a budget constraint. Link reinforcement in PSO-MLR has two purposes: maintaining network connectivity and reducing the worst-case congestion ratio. Numerical results show that PSO-M offers lower worst-case congestion ratios than the start-time optimization model, where link weights are obtained against the non-failure pattern assuming that multiple link failures are possible. The superiority of PSO-M strengthens as the average node degree of the network increases. Given a fixed budget, PSO-MLR allows the worst-case congestion ratio to be varied within a specific range. PSO-MLR can support a part of non-connected failure patterns to determine link weights, and so is a valuable enhancement of PSO-M.

This paper proposes a robust optimization model for probabilistic protection with multiple types of resources to minimize the required backup capacity for each type of resource against multiple random failures of physical machines in a cloud provider. If random failures occur, the required capacities for virtual machines are allocated to the preplanned backup physical machines, which are determined in advance. Probabilistic protection restricts the probability that the workload caused by failures exceeds the backup capacity by a given survivability parameter. We introduce three survivability parameters for central processing unit (CPU), memory, and the entire cloud provider considering both CPU and memory. By using the relationship between the three survivability parameters, the proposed model guarantees probabilistic protection for each resource, CPU and memory, and the entire cloud provider. By adopting the robust optimization technique, we formulate the proposed model as a multi-objective mixed integer linear programming problem. To deal with the multi-objective optimization problem, we apply the lexicographic weighted Tchebycheff method with which a Pareto optimal solution is obtained. We show that our proposed model reduces the average value between the backup capacity ratios of CPU and memory compared with the conventional model. A multi-objective simulated annealing (MOSA) and nondominated sorting genetic algorithm II (NSGA-II) are adopted to solve larger size problems. By using them, approximate solutions are obtained for larger size problems. In addition, we find that NSGA-II searches for solutions more effectively than MOSA, in our backup capacity allocation problem.

The availability schedule provides information on whether each network node is available at each time slot. The service interruptions caused by node unavailability marked in availability schedule can be suppressed if the functions are allocated according to the availability schedule. However, the given availability schedule may have gaps with the actual one and influence the VNF allocation. This paper proposes a robust optimization model to allocate virtual network functions (VNFs) in service function chains (SFCs) for time slots in sequence aiming to maximize the continuous available time of SFCs in a network with uncertain availability schedules by suppressing the interruptions caused by node unavailability marked in availability schedule and function reallocation. We formulate the problem as a mixed integer linear programming (MILP) problem over the given uncertainty set of the start time slot and period of unavailability on each node in the availability schedule. For solving the model in a practical time in a relative large size of network, we develop a heuristic algorithm. The numerical results show that the proposed model outperforms the baseline models under different levels of robustness in terms of the worst-case minimum number of the longest continuous available time slot in each SFC. The heuristic algorithm reduces the computation time with limited performance loss compared with the MILP approach. In the discussion, we introduce a constraint condition for the maintenance ability, which reduces the size of uncertainty set, and an extension for supporting more than one unavailability periods in the availability schedule on each node.

A suitable virtual network function (VNF) placement considering a node availability schedule extends service continuous serviceable time by suppressing service interruptions caused by function reallocation and node unavailabilities. However, function placement cannot avoid service interruptions caused by node unavailabilities. This paper proposes a primary and backup VNF placement model to avoid service interruptions caused by node unavailabilities by using backup functions. The considered backup functions have a period of startup time for preparation before they can be used and the number of them is limited. The proposed model is formulated as an integer linear programming problem to place the primary and backup VNFs based on the availability schedule at continuous time slots. We aim to maximize the minimum number of continuously available time slots in all service function chains (SFCs) over the deterministic availability schedule. We obverse that the proposed model considering the limited number of backup functions outperforms baseline models in terms of the minimum number of longest continuous available time slots in all SFCs. We introduce an algorithm to estimate the number of key unavailabilities at each time slot, which can find the unavailable nodes which are the bottlenecks to increase the service continuous available time at each time slot.

This paper proposes an optimization model to derive the virtual network function (VNF) allocation of time slots in sequence aiming to maximize the continuous available time of service function chains (SFCs) in a network. The proposed model suppresses service interruptions otherwise created by the unavailability of virtual machines (VMs) and the reallocation of VNFs. The proposed model computes VNF allocation in a series of time slots based on a VM availability schedule, which provides information on the availability of each VM in each time slot. We formulate the proposed model as an integer linear programming (ILP) problem with the goal of maximizing the minimum number of longest continuous available time slots in each SFC. We prove that the decision version of the VNF allocation problem (VNFA) is NP-complete. As the size of ILP problem increases, the problem is difficult to solve in a practical time. We develop a heuristic algorithm to solve the VNFA problem. Numerical results show that the proposed model improves the continuous available time of SFCs compared with existing models, which partially consider VM unavailability or VNF reallocation. We observe that the proposed model together with a consideration of routing reduces the path length of requests. The developed heuristic algorithm is faster than the ILP approach with a limited performance penalty.

This special section features the latest research contributions regarding the design and management of reliable networks. Reliability of communication infrastructure is a top priority for network operators. To ensure reliable network operation, new design and management techniques for reliable communications must be constantly devised to respond to the rapid network and service evolution. As a recent and relevant example, deployments of 5G communication networks will soon enter their second phase, during which the network infrastructure will require upgrades to support new Ultra-Reliable Low-Latency Communication (URLLC) services with availabilities of up to 6 nines to be guaranteed jointly with extremely low latencies. Even in the still preliminary vision of 6G communication networks, reliability is posed as one of the most critical requirements, as 6G networks will represent the communication platform of our future hyper-connected society, supporting essential services as smart mobility, e-health, and immersive environments with application in remote education and working, just to name a few. Similarly, disaster resiliency in communication networks is now attracting the attention of media, government and industry as never before (consider, e.g., the worldwide network traffic deluge to support remote working during the current Coronavirus pandemic). Luckily, several new technical directions can be leveraged to provide new solutions for network reliability as: increased network reconfigurability enabled by Software Defined Networking (SDN); integration/convergence of multiple technologies (optical, wireless satellite, datacenter networks); enhanced forms of data/service replication, supported by, e.g., edge computing; network slicing, used to carve highly-reliable logical partitions of network, computing and storage resources. These, and many others, technological transformations can be leveraged to enable next-generation high-reliability networks.

In 5G networks, multi-access edge computing (MEC) can be embedded into an access network (AN-MEC) and a core network (CN-MEC), which composes a two-tier MEC architecture for better scalability. In pre-Central Office Re-architected as a Data center (pre-CORD), AN-MECs are connected to a single but distant CN-MEC through Central Offices (COs). Disaggregation and virtualization of 5G network functions push CN-MEC into COs, which is known as post-CORD. Post-CORD has more CN-MECs closer to User Equipments than pre-CORD. In this work, we propose a scalable two-tier, multi-site, multi-server MEC architecture for pre-CORD and post-CORD. To adjust capacity and traffic allocation in such a distributed two-tier architecture, we integrate scaling and offloading with the objective of minimizing total capacity cost subject to the latency satisfaction percentage constraints, and solve the problem by Latency Aware Two-Phase Iterative Optimization (LA-TPIO). The results show that post-CORD with ten CN-MEC sites requires 30% less capacity than pre-CORD in satisfying 95% of URLLC traffic. Post-CORD utilizes about 48-77% less AN-MEC capacity than pre-CORD because post-CORD’s aggregated but close-enough CN-MEC sites are ideal for serving URLLC traffic. Under heavy hotspot traffic, post-CORD’s vertical and horizontal offloading percentages are 72% and 28%, respectively, while pre-CORD’s are 99% and 1%, which means post-CORD introduces more horizontal offloading because it has links between not only AN-MEC sites but also CN-MEC sites to accommodate hotspot traffic.

This paper proposes a multiple backup resource allocation model with a workload-dependent failure probability to minimize the maximum expected unavailable time (MEUT) under a protection priority policy. The workload-dependent failure probability is a non-decreasing function which reveals the relationship between the workload and the failure probability. The proposed model adopts hot backup and cold backup strategies to provide protection. For protection of each function with multiple backup resources, it is required to adopt a suitable priority policy to determine the expected unavailable time. We analyze the superiority of the protection priority policy for multiple backup resources in the proposed model; we provide the theorems that clarify the influence of policies on MEUT. We formulate the optimization problem as a mixed integer linear programming (MILP) problem. We provide a lower bound of the optimal objective value in the proposed model. We prove that the decision version of the multiple resource allocation problem in the proposed model is NP-complete. A heuristic algorithm inspired by the water-filling algorithm is developed with providing an upper bound of the expected unavailable time obtained by the algorithm. The numerical results show that the proposed model reduces MEUT compared to baselines. The priority policy adopted in the proposed model suppresses MEUT compared with other priority policies. The developed heuristic algorithm is approximately 106times faster than the MILP approach with 10-4performance penalty on MEUT.

Spectrally-spatially elastic optical networks (SS-EONs) have been considered nowadays to overcome the physical barrier and enhance the transport capacity, where enhancing spectrum utilization while satisfying inter-core and inter-mode crosstalks is always challenging. This paper proposes a priority-based crosstalk-avoided core, mode and spectrum allocation scheme in SS-EONs, which enhances resource utilization while satisfying both constraints of inter-core crosstalk and inter-mode crosstalk. The proposed scheme creates different groups of cores and modes and assigns a priority to each of them. Core and mode are selected for serving lightpath requests based on their priority order. We define an optimization problem for routing, modulation assignment, spectrum, core, and mode allocation (RA-SCMA) in SS-EONs considering both constraints of inter-core crosstalk and inter-mode crosstalk simultaneously. The optimization problem is formulated as an integer linear programming problem. We prove that the decision version of RA-SCMA is NP-complete. We present crosstalk-avoided core-mode-spectrum allocation considering a dynamic scenario. Numerical results indicate that the proposed scheme reduces the blocking probability in SS-EONs and allows up to 40% increased traffic loads by utilizing the crosstalk-avoided unutilized slots compared to the conventional scheme that adopts a core-mode-spectrum first fit policy.

Service chaining is a method for providing desired network services to users by concatenating virtualized network functions (VNFs) in the network. There have been studies on service chain provisioning models that relax the visit order of VNFs and routing constraints. These models make it difficult to obtain an optimal solution in a practical time due to the huge number of decision variables associated with the problem. A heuristic approach that obtains a nearly-optimal solution within a practical time is needed. This paper proposes a column generation based heuristic algorithm for the service chain provisioning problem that relaxes the VNF visit order and routing constraints. The proposed algorithm divides the problem into a VNF placement problem and a routing problem and applies the column generation technique to solve the latter. Numerical results show that the proposed algorithm shortens the computation time compared to directly solving the original integer linear programming problem in exchange for some increase in the cost for VNF placement and link utilization.

This paper proposes a multiple backup resource allocation model with a workload-dependent failure probability to minimize the maximum expected unavailable time (MEUT) under a protection priority policy. The workload-dependent failure probability is a monotonically increasing function which reveals the relationship between computing workload and failure probability. The proposed model adopts hot backup and cold backup strategies to provide protection. The cold backup strategy is a protection strategy, in which the requested loads of backup resources are not processed as active workloads before failures occur to reduce resource utilization with the cost of long recover time. The hot backup strategy is a protection strategy, in which the backup resources execute at the same time with functions to recover promptly with the cost of high workload. For protection of each function with multiple backup resources, it is required to adopt a suitable priority policy to determine the expected unavailable time. We analyze the superiority of the protection priority policy for multiple backup resources in the proposed model and provide the theorems that clarify the influence of policies on MEUT. The numerical results show that the proposed model reduces MEUT compared with the single backup model in which each function is protected by only one server without protection priority of servers. The priority policy adopted in the proposed model specifying that the server which adopts the hot backup strategy has higher priority than that with the cold backup strategy for multiple backup resources suppresses MEUT compared with other priority policies.

This paper presents two strategies to obtain an assignment of backup servers to network functions of middleboxes when each backup server can recover a half of the functions which it protects at the same time. In the previous work, there are approaches to obtain an assignment only when each backup server protects two functions and recovers one of them at the same time. Therefore, we present two strategies to expand the cases where an assignment can be obtained by utilizing the previous approaches. The basic ideas of our two strategies are dividing each server into a set of small servers that protects two functions and recovers one of them at the same time, obtaining an assignment with them, and combining them. In the process of obtaining an assignment with our presented two strategies, there is a constraint to avoid impairing the capabilities of backup servers. Our two strategies incorporate this constraint before and after obtaining an assignment with the divided small servers, respectively. We define six survival probabilities regarding our two strategies and analyze their relationships. Then, we derive two theorems to consider when our strategy can obtain an assignment that satisfies the constraint. Based on the theorems, we analyze properties of our strategies and the relationship between the different survival probabilities. Numerical results show that one of our strategies provides the higher survival probability than the other one for all the settings that we examine.

This paper proposes a virtual network function (VNF) placement model for service chaining that minimizes the cost of using computation resources when no failure occurs while guaranteeing recovery against any single facility node failure within the recovery time objective (RTO) defined for each service. The proposed model adaptively allocates computation resources to each service under its RTO constraint. The proposed model introduces two sharing methods of computation resources among multiple service chains. The first method allows sharing a virtual machine (VM) where a VNF is scheduled to run after a failure, which contributes to suppressing the number of VMs reserved in preparation for a failure. The second method allows sharing computation capability used for VMs, which prevents unnecessary VNF scale-up that requires additional computation resources. A simulation study verifies that the proposed model reduces the cost of using computation resources compared to comparative models.

This paper proposes a distributed server allocation model with the preventive start-time optimization against a single server failure. The proposed model preventively determines the assignment of servers to users under each failure pattern to minimize the largest maximum delay among all failure patterns. We formulate the proposed model as an integer linear programming problem. We prove the NP-completeness for the considered problem. The numerical results reveal that the proposed model reduces the largest maximum delay compared to one baseline; it avoids instability caused by the unnecessary disconnection, which frequently occurs in the other baseline.

This paper proposes a preventive priority setting model to handle load balancing against multiple controller failures in software defined networks. For each switch, a set of controllers can control it, where only one master controller controls the switch and others are slave controllers. We introduce a priority for each controller to become the master controller. At any time, the controller which does not fail and has the highest priority works as the master controller. Once the existing master controller fails, the assignment of new master controller is automatically obtained according to the priority setting to promptly recover the control. The priority set is decided at the network operation start time to minimize the maximum utilization ratio among controllers in the worst-case of failure patterns. We formulate the proposed model in two different forms, which are an integer linear programming formulation and a min-max formulation. We prove that the preventive priority setting problem is NP-hard. A heuristic algorithm is introduced based on the min-max formulation. The numerical results reveal that the proposed model obtains the maximum utilization ratio comparable to those obtained by two baselines; it provides a faster recovery compared to one baseline and reduces the instability of network operation compared to the other.

Real-time prediction of communications (or road) traffic by using cloud computing and sensor data collected by Internet-of-Things (IoT) devices would be very useful application of big-data analytics. However, upstream data flow from IoT devices to the cloud server could be problematic, even in fifth generation (5G) networks, because networks have mainly been designed for downstream data flows like for video delivery. This paper proposes a framework in which a software defined network (SDN), edge server, and cloud server cooperate with each other to control the upstream flow to maintain the accuracy of the real-time predictions under the condition of a limited network bandwidth. The framework consists of a system model, methods of prediction and determining the importance of data using machine learning, and a mathematical optimization. Our key idea is that the SDN controller optimizes data flows in the SDN on the basis of feature importance scores, which indicate the importance of the data in terms of the prediction accuracy. The feature importance scores are extracted from the prediction model by a machine-learning feature selection method that has traditionally been used to suppress effects of noise or irrelevant input variables. Our framework is examined in a simulation study using a real dataset consisting of mobile traffic logs. The results validate the framework; it maintains prediction accuracy under the constraint of limited available network bandwidth. Potential applications are also discussed.

Network function virtualization paradigm enables us to implement network functions provided in middleboxes as softwares which run on commodity servers. This paper proposes an unavailability-aware backup allocation model with shared protection for middleboxes with comprehensively considering the failure, repair, and recovery behaviors of functions and backup servers. Multiple functions can share the backup resources on the backup server. The proposed model aims to find the assignment of backup servers to functions to minimize the maximum unavailability among functions. The multiple situations of failure, repair, and recovery of functions and backup servers lead to several different states for each function. The unavailability of function is estimated through analyzing all states that a function can be in. To compute the unavailability of middlebox for a given backup allocation, an analytical approach is developed based on the queueing theory. With the analytical approach, we introduce a simulated annealing heuristic to solve the backup allocation problem. The results reveal that, compared to a baseline model, the proposed unavailability-aware model reduces the maximum unavailability 11% in average in our examined scenarios.

A virtual network function allocation model to maximize continuous available time of service function chains was introduced in our previous work. The performance of this model needs to be evaluated on network devices. It is time-consuming and costly to deploy functions with real network devices. Existing simulation tools require powerful computation capability, which limits the usable cases. We implement a network service header based service function chain application which can be cooperated with the model. Demonstration validates that the application allocates functions by using the allocation from the model automatically and runs service function chains correctly.

This paper proposes an optimization model for the network service mapping and scheduling problem with uncertain processing time. We model processing time uncertainty through Γ-robustness approach, which provides different degrees of robustness against processing time uncertainty. We formulate the problem with the objective to minimize the worst-case makespan over the given uncertainty set. A heuristic is presented to tackle the problem. The numerical results show that the proposed model outperforms the conventional model with deterministic parameters in terms of worst-case makespan.

This paper proposes a backup network design scheme that can determine backup link capacity in practical time. The proposed scheme suppresses the required backup link capacity while providing a guaranteed level of recovery against multiple independent link failures. The conventional scheme is based on robust optimization and suffers from the problem of overestimating the backup link capacity. The proposed scheme addresses the overestimation problem by computing the probabilistic distribution function of required backup link capacity in polynomial time. We formulate the backup network design problem with the proposed scheme as a mixed integer linear programming problem to minimize the total required backup link capacity. We prove that the decision version of backup network design problem is NP-complete. Given that network size will continue to increase, we introduce a heuristic approach of simulated annealing to solve the same problem. Numerical results show that the proposed scheme requires less total backup link capacity than the conventional scheme based on robust optimization.

Network function virtualization enables network operators to implement network functions in a software-oriented manner and makes network services (NSes) provisioning much simpler. This paper proposes an optimization model to schedule delay sensitive NSes with deadlines allowing resource sharing and preemption. Unlike conventional NS scheduling models with static resource allocation for virtualized network function (VNF) instances, the proposed model ensures that VNF instances deployed on the same node share computation resources of the node and are able to scale up/down to change their process rate at runtime. NSes mapped to the same VNF instance of the same node share computation resources of the VNF instance and are able to be processed in parallel by the VNF instance. Preemption is allowed, which means that rescheduling the order of NS processing at runtime is possible and the process duration of each function of an NS is allowed to be discrete. We formulate the proposed model as an integer linear programming problem to maximize the number of admissible NSes. Due to the complexity of the problem, we develop a genetic algorithm to solve it efficiently. We evaluate the proposed model with conventional models in the static and dynamic scenarios. The numerical results show that the proposed model outperforms conventional models in terms of acceptance ratio in both static and dynamic scenarios.

Software-defined network (SDN) is a network that the centralized SDN controller stores flow entries in the flow table of each SDN switch and controls packet flows as instructed by the stored flow entries. When a multicast service is provided in an SDN, the SDN controller stores a multicast entry dedicated for a multicast group in each SDN switch. It is necessary to suppress the number of flow entries required to set up a multicast tree due to the limited capacity of the flow table. In a conventional research, a multicast routing model that suppresses the number of multicast entries in one multicast request by replacing a part of them with unicast entries has been devised. However, since this conventional model individually determines a multicast tree route for each request, unicast entries configured for the same receiver are distributed in various SDN switches when multiple multicast services are requested. As a result, there is still the possibility of improving the reduction of the number of flow entries. In this paper, we propose a multicast routing model for multiple multicast requests that minimizes the number of flow entries. This proposed model determines multiple multicast tree routes simultaneously so that a unicast entry configured for the same receiver and stored in the same SDN switch is shared by multicast trees. We formulate the proposed model as an Integer Linear Programming (ILP) problem. Numerical results show that the proposed model reduces the required number of flow entries compared to the conventional model.

This paper proposes a server selection scheme in successive participation scenario based on the segmentation of participating-domain to suppress the latency. In the proposed scheme, the users participate for server selection one after another. The proposed scheme determines a recommended server, and a new participating user selects the recommended server first. Recommended servers are determined in advance at the condition that participating users exist in all regions in users' participation domain. A recommended server is determined for each divided region to minimize the latency when there exists one user at each region. The new participating user selects the recommended available server, where the user is located. We formulate an integer linear programming problem to determine the recommended servers for the proposed scheme. We use the outcome of the recommended server finding process as the input for server selection. Numerical results express that smaller latency is obtained using the proposed scheme compared to the conventional greedy based server selection scheme, by employing some additional computations for finding the recommended servers before participating of users.

Real-time prediction of spatial information has attracted a lot of attention. Machine learning enables us to provide real-time prediction of spatial information such as road traffic by using aggregated sensor data. The amount of mobile traffic is forecasted to increase exponentially, thereby causing serious transmission delays when traffic loads are heavy. If a part of the data used for predicting spatial information in real time does not arrive on time, the prediction accuracy degrades because the prediction is done without the missing data. A utility-based scheduling technique has been suggested as a way of prioritizing such delay-sensitive data. However, no study has not addressed the utility-based scheduling for the real-time prediction of spatial information. Therefore, this paper proposes a scheme that enables modeling the utility function for real- time prediction of spatial information. The scheme is roughly composed of two steps: the first creates training data from original time-series data and a machine learning model using the data, while the second models the utility function using the feature selection method in the learning model. Feature selection method enables extracting the importance of data in terms of how much the data contributes to the prediction accuracy. This paper assumes the road traffic prediction as a scenario and shows the utility function modeled by the proposed scheme using real spatial datasets. A numerical study demonstrates how the model of the utility function works effectively in prioritizing data for real-time prediction in terms of accuracy.

This paper presents three approaches to solve the problem of finding the optimal backup resource assignment which maximizes the survival probability of network functions of middleboxes. In the previous work, no mathematical model to solve this problem is provided, so we formulate the problem as a mixed-integer linear programming (MILP) problem as the first approach. Formulating this MILP problem includes some special steps, which are not considered in the previous work. The MILP problem is not always solved in a practical time when the problem size becomes large. Then, we develop two heuristic approaches by replacing the objective of the original MILP problem relying on the idea of balancing the failure probabilities of functions of connected components. Numerical results show that our two developed heuristic approaches improve the survival probability from a conventional heuristic algorithm in some cases and reduce computation time compared to obtaining the optimal solution. Furthermore, one of our developed heuristic approaches provides exactly the optimal solution with shorter computation time compared to the time solving the original MILP problem in a special case.

In this paper, we propose a virtual network function placement and routing model for multicast service chaining based on merging multiple service paths (MSC-M). The multicast service chaining (MSC) provides a multicast path, which connects a source node and multiple destination nodes, and virtual network functions (VNFs) are placed on the path so that users on the destination nodes receive their desired services. The conventional MSC model configures multicast paths for services, each of which has the same source data and the same set of VNFs in a predefined order. In the MSC-M model, if paths of different services carry the same data on the same link, these paths are allowed to be merged into one path at that link, which improves the utilization of network resources. The MSC-M model determines the placement of VNFs and the route of paths so that the total cost associated with VNF placement and link usage is minimized. The MSC-M model is formulated as an integer linear programming (ILP) problem. In the ILP problem, data flows whose source data is the same and which already passed the same subset of VNFs belong to the same group. A part of paths of different services which carry data flows belonging to the same group are allowed to be merged into one path. Numerical results show that the MSC-M model reduces the total cost by 28.7% at a maximum compared to the conventional MSC model.

The performance of packet processing applications is dependent on memory accesses speed of network systems. Table lookup requires fast memory accesses and is one of the most common processes in various packet processing applications, which can be a dominant performance bottleneck. Therefore, in Network Function Virtualization (NFV)-aware environment, on-chip fast cache memories of a CPU of general-purpose hardware become critical to achieve high performance packet processing over tens of Gbps. In addition, multiple types of applications and complex applications are executed in the same system simultaneously in carrier network systems, which require the capacity of cache memories as well. In this paper, we propose a packet processing architecture that utilizes interleaved 3 Dimensional (3D)-stacked Dynamic Random Access Memory (DRAM) devices as off-chip Last Level Cache (LLC) in addition to several levels of dedicated cache memories of each CPU core. Entries of a lookup table are distributed in every bank and vaults to utilize both bank interleaving and vault-level memory access parallelism. Frequently accessed entries in 3D-stacked DRAM are also cached in dedicated on-chip cache memories of each CPU core. The evaluation results show that the proposed architecture reduces the memory access latency by 57 % and increases the throughput by 100 % with reducing blocking probability about 10 % compared to the conventional architecture with common on-chip LLC. These results indicate that 3D-stacked DRAM can be practical as off-chip LLC in parallel packet processing running on multiple CPU cores simultaneously.

This paper proposes an optimization model to schedule network services (NSes) in virtual networks with resource sharing and preemption. Inefficient NS scheduling can severely degrade the acceptance ratio of arriving NSes of the network. Conventional NS scheduling models do not consider sharing computational resources of a node among different virtual network function (VNF) instances deployed on this node. In the proposed model, NSes mapped to the same VNF instance on the same node share computational resources of the VNF instance, and VNF instances deployed on the same node share computational resources of the node. The proposed model allows preemption, which means that rescheduling the process order of NSes in runtime is possible and the process duration of each function of an NS is allowed to be discrete. We formulate the proposed model as an integer linear programming problem to maximize the number of admissible NSes. The numerical results show that the proposed model outperforms conventional models in terms of the acceptance ratio of arriving NSes.

This paper proposes a master and slave controller assignment model against multiple controller failures in software defined network with considering propagation latency between switches and controllers. In our model, a controller can be assigned to multiple switches, and the survivability of each switch is guaranteed to a certain degree by assigning multiple controllers to it. We define the average-case expected propagation latency, the worst-case expected propagation latency, and the expected number of switches within a propagation latency bound, as three different objectives to be optimized, which lead to three different problems, in this paper. We formulate the proposed master and slave controller assignment model with different goals as three mixed integer linear programming problems. Results show that the optimal assignments vary for different problems. A greedy algorithm with polynomial time complexity is introduced to solve the same optimization problems. We evaluate the performance of introduced greedy algorithm compared with the optimal value in one of the problems, which minimizes the average-case propagation latency. The numerical results reveal that the computational time of running the greedy algorithm to obtain a solution is about 10-3times compared to that of solving the mixed integer linear programming problem; the obtained objective value is about 1.00324 times of the optimal value in average in our examined scenarios.

The machine-to-machine (M2M) service network platform that accommodates and controls various types of Internet of Things (IoT) devices has been presented. This paper investigates the program file placement strategies for the M2M service network platform which achieve low blocking ratio of new task requests and accommodate as many tasks as possible in the dynamic scenario. We present four strategies to arrange the program file placement, which differ in the objective function, the computation method, and the timing of rearrangement of whole program file placement. The simulation results show that a strategy based on solving a mixed-integer linear programming (MILP) model achieves the lowest blocking ratio and the highest number of completed tasks within a certain time period. In the case that the MILP-based strategies are intractable due to the computation time, a heuristic algorithm-based strategy which basically determines the program file placement for only a newly requested task achieves low blocking ratio.

The lightpath provisioning time or releasing time is adversely affected by the time that optical amplifiers require to adjust to a newly added or terminated signal power. This shortcoming is particularly true with multi-core erbium-doped amplifiers (EDFAs), as multi-core transient-suppressed EDFAs are unavailable at the current time. This paper proposes a fiber span power management scheme based on dummy wavelength signals that are used to shorten the lightpath provisioning and releasing times in multi-core fiber networks. With the shorter time of lightpath provisioning and releasing procedures, the total time that is required to reserve wavelengths in the system is decreased, which means that network resources are used more efficiently. As a result, the blocking performance and average waiting time in the system are improved. To evaluate the performance of the proposed scheme, this paper introduces both analytical model and simulation study. In the introduced model, the ratio of the number of activating and activated dummy wavelengths to the number of dummy wavelengths in each span is considered in the range between 0 and 1. The analysis reveals that the performance of the proposed scheme depends on α, which is the ratio of the number of dummy wavelengths to the number of dummy and lightpath wavelengths in each span, and there exists a point of α where the blocking probability becomes minimum. We further observe that the proposed scheme outperforms the conventional approaches in terms of blocking probability and average waiting time, as traffic loads increase. Finally, we provide the direction on how our introduced model can be considered for a network with multi-span routes.

Network function virtualization paradigm enables us to implement network functions provided in middleboxes as softwares that run on commodity servers. This paper proposes a backup resource allocation model for middleboxes with considering both failure probabilities of network functions and backup servers. A backup server can protect several functions; a function can have multiple backup servers. We take the importance of functions into account by defining a weighted unavailability for each function. We aim to find an assignment of backup servers to functions, where the worst weighted unavailability is minimized. We formulate the proposed backup resource allocation model as a mixed integer linear programming problem. We prove that the backup resource allocation problem for middlebox with importance is NP-complete. We develop three heuristic algorithms with polynomial time complexity to solve the problem. We analyze the approximation performances of different heuristic algorithms with providing several lower and upper bounds. We present the competitive evaluation in terms of deviation and computation time among the results obtained by running the heuristic algorithms and by solving the mixed integer linear programming problem. The results show the pros and cons of different approaches. With our analyses, a network operator can choose an appropriate approach according to the requirements in specific application scenarios.

This work proposes a defragmentation scheme using reroutable backup paths in toggled-based quasi 1+1 path protected elastic optical networks to enhance the efficiency of defragmentation and suppress the fragmentation effect. The proposed scheme allows both reallocation of spectrum slots of backup paths and rerouting of backup paths. By using the path exchanging approach in the proposed scheme, the primary paths become the backup path while the backup path becomes the primary path. This allows to utilize the advantages of defragmentation in both primary and backup paths. Considering rerouting and path exchanging, we present to the key idea to formulate the proposed scheme as an integer linear programming (ILP) problem. A heuristic algorithm is introduced to solve the problem for large networks, when ILP is not tractable. For a dynamic traffic scenario, an approach that suppresses the fragmentation considering rerouting and path exchanging operations is presented. The numerical results indicate that the blocking probability using the proposed scheme is suppressed compared to the conventional scheme.

Recently, the demand for realtime and accurate prediction of mobile traffic has been growing in traffic engineering and dynamic resource allocation that work to handle increased mobile data traffic. However, most conventional prediction techniques assumed that traffic logs at every unit time at every base station are perfectly available. This assumption is critical in realtime mobile traffic prediction because the volume of traffic log data collected at base stations is huge and they compete bandwidth with normal user application traffic when they are sent from base stations to the server that performs prediction. Therefore, in realtime mobile traffic prediction, we should consider the condition in which the bandwidth ensured for forwarding traffic log data is limited. In this paper, we propose a method that prioritizes traffic log data in the basis of the contribution to prediction accuracy; each base station sends more important traffic log data to the server with higher priority. The importance of each data entry of traffic log data means how much prediction accuracy would degrade if the entry is missing. The proposed method enables us to reduce the volume of traffic log data sent from base stations to the server while maintaining prediction accuracy at the sufficient level. Our simulation study using a real dataset of mobile-traffic measurement validates our method in terms of prediction accuracy under the limitation of available traffic log data.

This paper shows how to design, within practical time constraints, a backup network that suppresses the required resources while providing a guaranteed level of recovery against multiple link failures. The conventional scheme based on robust optimization has the problem of overestimating the backup link capacity. The backup network design scheme proposed herein computes the probabilistic distribution function of required backup link capacity in polynomial time, and so addresses the optimization problem of minimizing the total backup network capacity. For large networks, we introduce the heuristic approach of simulated annealing that adopts our approach to computing backup link capacity. Numerical analyses show that the proposed scheme requires less total backup network capacity than the conventional scheme based on robust optimization.

This paper proposes a backup resource allocation model that provides a probabilistic protection for primary physical machines in a cloud provider to minimize the required total capacity. When any random failure occurs, workloads are transferred to preplanned and dedicated backup physical machines for prompt recovery. In the proposed model, a probabilistic protection guarantee is introduced to prevent the cloud provider from capacity overbooking. We apply robust optimization in our model to formulate the backup resource allocation problem as an integer linear programming problem. A simulated annealing heuristic is adopted to solve the same optimization problem when the cloud provider is large. Finally, the results reveal that the required backup capacity depends on the reliability of primary physical machines. Specifically, the more the resources in primary physical machines share backup capacity when the failure probabilities of primary physical machines are sufficiently small, the less capacity is required for backup resource allocation.

Emergence of new network services such as Internet of Things (IoT) and edge computing accelerates the increase of traffic volume, the number of connected devices and the diversity of communication. Next generation carrier network infrastructure should be much more scalable and adaptive to rapid increase and divergence of network demand with much lower cost. More virtualization-aware, flexible and inexpensive system based on general-purpose hardware is necessary to transform traditional carrier network into more adaptive, next generation network. In this paper, we propose a carrier-scale packet processing system which utilizes 3 Dimensional (3D)-stacked Dynamic Random Access Memory (DRAM) device. The proposed system augments memory access concurrency by leveraging vault-level parallelism and bank interleaving of 3D-stacked DRAM. The system uses hash-function-based distributor of memory requests to each set of vault and bank which accommodates a portion of original carrier-scale huge tables. We introduce an analytical model for the system. The evaluation result shows that our proposed system can achieve more than 100 Gbps in carrier-scale packet processing where main memory accesses are inevitable since tiny CPU cache memory is insufficient to accommodate huge tables. Our analytical model is independent of specification of a particular device, which can be applied to any DRAM systems.

In this paper, we propose an optimization model for designing multiple network functions virtualization (NFV)-based campus area networks (CANs). Organizations, such as universities and research institutions, have their own campus information and communication technology (ICT) equipment, and it is desired that this equipment be moved to NFV/cloud data centers of high reliability and resiliency. However, NFV-based CAN is not affordable because the data transmission cost is higher with a public cloud. One solution is for multiple organizations to procure NFV/cloud data center resources together. By doing so, the cost of these resources will be reduced. There are planning issues to solve when choosing optimal NFV/cloud sites in order to make progress on this approach. The proposed model minimizes the total network cost incurred by multiple organizations including the wide area network cost. It is formulated and analyzed by using mixed integer liner programming. The effect of cost minimization was evaluated in a ladder network, and the cost reduced up to 50%. This reduced cost will encourage organizations to deploy NFV-based CANs.

We propose an optimization model for designing multiple network functions virtualization (NFV)-based campus area networks (CANs). Organizations, such as universities and research institutions have their own campus information and communication technology equipment, but many would like to move this equipment to NFV and cloud data centers for improving reliability and resiliency. However, NFV-based CAN is not affordable for them, because costs are higher with a cloud. One solution is for multiple organizations to procure NFV and cloud data center resources together. By doing so, their individual costs of using these resources will be reduced. To make progress on this approach, there are planning issues to resolve when choosing optimal NFV and cloud data center locations. The proposed model minimizes the total network costs incurred by the organizations, including the wide area network cost and data synchronization costs for recovery from faults at data centers and the various subcampus network configurations of legacy CANs. The model is formulated and analyzed by using mixed integer linear programming. The effect of cost minimization is evaluated in a ladder network and an actual network, SINET5, and it is found that the costs can be reduced by up to 63%. The calculation times of this model under practical conditions are short and the model will be useful in practice. It is also shown that the cost of fault recovery can be suppressed. These results will encourage organizations to deploy NFV-based CANs.

We propose a virtualization architecture for multi-campus information and communication technology (ICT) equipment with integrated cloud and NFV capabilities. The aim of this proposal is to migrate most ICT equipment on campus premises into cloud and NFV platforms. Adopting this architecture would make most ICT services secure and reliable and their disaster recovery (DR) economically manageable. We also analyze a cost function and show the cost advantages of this proposed architecture, describe implementation design issues, and report a preliminary experimentation of NFV DR transaction. This architecture would encourage academic institutes to migrate their own ICT systems located on their premises into cloud environments.

Optical networks are well suited to support massive data exchanges between data centers. Elephant traffic flows can be routed over provisioned and dedicated lightpaths (optical flows) while other (mice) flows, which are routed by electronic switches, are unaffected. In some solutions, Wavelength-Selective Switches (WSSs) are employed in the optical nodes to individually route the lightpath towards its destination. WSSs take time to be switched and delay the lightpath setup time. In this paper, the authors compare three service policies for WSS devices aiming to reduce the lightpath setup and tear down times. The conventional service policy assumes that each setup (tear down) request is handled individually. Two other service policies assume that groups of setup (tear down) requests are handled together by the WSS. These policies are implemented in a discrete event simulator, which is used to estimate the end-to-end lightpath setup and tear down time across an arbitrary mesh network. Simulation results show that group service policies outperform the conventional policy at high loads. The grouping policies are useful to reduce the lightpath setup time especially in the presence of lightpaths that are frequently set up and have relatively short holding time (short duration of elephant-optical flows).

Optical switching technologies offer a cost-and power-efficient approach for handling the DataCenter Network (DCN) oversubscription problem. We propose a hybrid DCN architecture named Optical Switch in the Middle (OSM), which offers increased flexibility (when compared to prior hybrid architectures) for supporting multiple simultaneous high-speed TOR-to-TOR paths through an Optical Circuit Switch (OCS) and a core-level Electrical Packet Switch (EPS). A multilayer SDN controller supports advanced-reservation scheduling of optical circuits, and the integration of storage in the core EPS increases the usage rate of optical circuits. To effectively use the OSM architecture, we propose four modifications to Hadoop, and illustrate the potential of this architecture for achieving higher compute-resource utilization while simultaneously offering users shorter job completion times.

In a high-speed backbone network, the failure of a network link may cause large data losses, so it is necessary to reserve spare network resources for faster recovery. The conventional protection methods to reserve backup routes do not consider the failure probability of each network link, so the same amount of network resources for the backup route are needed regardless of the failure probability of network links. This leads a decrease in the number of connections that can be accepted into the network. This paper proposes a routing and capacity allocation method that guarantees the expected value of allocated capacity. We formulate a mixed integer liner programming model for the proposed method. We conduct simulations to study the advantage of the expected capacity guaranteed routing over the conventional routing method in terms of bandwidth blocking probability. The results show that the proposed method reduces the bandwidth blocking probability to about 1/3 as compared to that of the conventional path protection method.

In elastic optical networks (EONs), a major obstacle to using the spectrum resources efficiently is the spectrum fragmentation. In the literature, several defragmentation approaches have been presented. For 1+1 path protection, conventional defragmentation approaches consider designated primary and backup paths. This exposes the spectrum to fragmentations induced by the primary lightpaths, which are not to be disturbed in order to achieve hitless defragmentation. This paper proposes a defragmentation scheme using path exchanging in 1+1 path protected EONs. We exchange the path function of the 1+1 protection with the primary toggling to the backup state, while the backup becomes the primary. This allows both lightpaths to be reallocated during the defragmentation process, while they work as backup, offering hitless defragmentation. Considering path exchanging, we define a static spectrum reallocation optimization problem that minimizes the spectrum fragmentation while limiting the number of path exchanging and reallocation operations. We then formulate the problem as an integer linear programming (ILP) problem. We prove that a decision version of the defined static reallocation problem is NP-complete. We present a spectrum defragmentation process for dynamic traffic, and introduce a heuristic algorithm for the case that the ILP problem is not tractable. The simulation results show that the proposed scheme outperforms the conventional one and improves the total admissible traffic up to 10%.

This paper analyzes the computational time complexity of the allocation problem for data processing functions among multiple users and distributed servers in the distributed processing communication scheme for a real-time network application. In the distributed processing communication scheme, the application is processed on a data processing function in the distributed servers in order to minimize the delay time. We prove that the allocation problem for data processing functions among multiple users and distributed servers is an NP-complete problem.

A flexible orthogonal frequency-division multiplexing optical network enables to change bandwidth flexibly by changing the number of sub-carriers. We consider that users request to dynamically change the number of sub-carriers. In this context, the network resources can be used more efficiently. The dynamic bandwidth change needs a certain time. Service centric resource allocation must be considered in terms of waiting time to change the number of sub-carriers. If the user demands drastically increase such as a disaster priority telephone service, a waiting time for bandwidth changes is not tolerated because emergency is time-critical. This is caused by a chain-change of bandwidth such as a multiple-car pileup. This paper proposes a grouped elastic spectrum allocation scheme to satisfy the tolerable waiting time of the service in an optical fiber link. Spectrums are grouped to restrict a waiting time in the proposed scheme. In addition, the proposed scheme determines a margin bandwidth with neighbor spectrums to prevent frequent reallocation by estimating real traffic behavior. Numerical results show that a required bandwidth can be saved with satisfying waiting time constraints. Additionally measurement granularity is discussed.

This paper proposes a source-based high-speed wavelength-path protection scheme configuring a tree-shaped backup path in WDM networks. In the proposed scheme, a failure-detecting node on a primary path starts protection by sending a failure notification to a source node of the primary path through the tree-shaped backup path. Intermediate nodes on the backup path and the source node which receive the failure notification switch to the backup path autonomously. Therefore, the proposed scheme provides high-speed protection. The simulation results revealed that the proposed scheme reduces the failure-recovery time compared with the conventional protection schemes.

Elastic Optical Networks (EONs) have the potential to improve the fiber spectrum utilization by allocating spectrum resources to multiple traffic requests proportionally to the amount of carried traffic. However, achieving high spectrum utilization in this elastic scenario is hindered by the resulting spectrum fragmentation. A number of studies have addressed and made attempts to mitigate spectrum fragmentation. Most of these studies are based on simulation techniques and target the overall blocking probability experienced by the offered traffic requests due to the lack of available spectrum resources. Some studies have also shown that blocking probability in EON can be uneven, i.e., high-rate circuit requests are more likely to be blocked when compared to low-rate requests due to the shortage of contiguously available spectrum resources. The contribution of this paper is to extend an existing Markov Chain (MC) model previously proposed by the authors to quantify blocking probability in a two-service elastic fiber link. The model extension accounts for a self-limited and partial sharing of the fiber spectrum to accommodate the two types of service. The MC model is used to quantify both the blocking probability and its fairness across the two types of service, documenting how the EON uneven blocking behavior can be significantly mitigated by performing partial (as opposed to full) sharing of the fiber spectrum.

This paper proposes a technique to split a traffic in a software-defined network using a meter table. In general case, a different differentiated services code point (DSCP) number in a packet header is used to classify network traffic and provide quality of service (QoS). The traffic splitting technique uses a DSCP number as a split parameter. Once a packet exceeds a predefined traffic rate, the DSCP value of the packet is changed. The packets with different DSCP number are sent out to the neighbor switches with different output ports.

This paper aims to design a Hadoop system and evaluates the performance of a task allocation scheme. The task allocation scheme splits each job into tasks using an appropriate splitting ratio, and assigns tasks to slave servers based on server processing performance and network resource availability. We experimentally evaluate the performance of the scale out of the task allocation scheme with five machines. We focus on the configuration of jobtracker and tasktracker in Hadoop. In cases with heterogeneous Hadoop clusters, we distribute task blocks to high-capability slaves with proportionally larger-sized tasks than to low-capability slaves. We create an environment in which high-capability slaves perform more work than low-capability slaves. The experimental testbed results indicate that the task allocation scheme is effective.

This paper introduces a green and robust optimization scheme based on hose model with bound of link traffic (HLT), in order to achieve power savings in the networks with traffic uncertainty. Most of the studies on green communications nowadays are based on estimates of real traffic matrix. However predicting the traffic matrix is a difficult task for network operators. Further, these models may not be fully applicable in a context where the traffic often fluctuates. By using HLT, the knowledge of the exact traffic information is not required. The traffic is specified by the total outgoing and incoming amount at each node and the total traffic going through each link. We formulate the problem as a mixed integer linear programming (MILP) problem, with an objective to reduce the flow through each link and allow the links to be put to sleep mode. We develop a heuristic to mitigate the limitations of the MILP formulation. Simulation results show that green HLT, while being robust to traffic uncertainty, achieves power efficiency comparable to the models where the knowledge of the traffic information is required.

This paper proposes a MIMO system where the base station (BS) acquires quantized phase-only (PO) measurements of the complex baseband signal by our introduced stage-wised phase quantizer. PO-MIMO requires only one-bit ADCs for data sampling, so it successfully overcomes the ADC bottleneck that appears when the signal bandwidth is extremely wide. We construct a PO generalized approximate message passing (POG-AMP) algorithm for solving the linear mixing problem with quantized or unquantized phase measurements. POG-AMP has low computational complexity, exploits the signal prior statistical distribution, and handles the nonlinear distortions exerted on the measurements (e.g., losing magnitude and quantization). Then, POG-AMP is successfully applied to construct practical channel estimator and multiuser detector for PO-MIMO. Numerical results show that the POG-AMP estimator (POG-AMPE) and POG-AMP detector (POG-AMPD) are robust to the phase-quantization loss. POG-AMPE acquires high-quality channel side information at the receiver (CSIR), and POG-AMPD is robust to the CSIR errors when the BS antennas are massive enough. By introducing moderately more BS antennas, PO-MIMO with phase measurements even performs similarly to MIMO with full measurements containing both magnitude and phase. In order to maximize the transmit energy-efficiency, the lengths of the channel training sequences should be gradually increased with the increase of the channel coherence time. Antenna correlations at the BS degrade the convergence and bit-error rate performances of POG-AMPD, but can be handled by the analog spatial filtering technique.

This paper proposes a multiuser MIMO system with both full measurements and one-bit magnitude observations, which are acquired by several linear inphase-and-quadrature (IQ) structured radio frequency (RF) chains and massive one-bit envelope chains, respectively. The total circuit power and cost are not increased significantly, since the added one-bit envelope chains have low power and low cost. Channel side information on the one-bit envelope chains is acquired by sharing the IQ-structured RF chains and executing a channel calibration operation. Two multiuser detectors are constructed based on the semidefinite relaxation (SDR) and approximate message passing (AMP). The one-bit magnitudes are interpreted as inequality constraints in the SDR detector, and exploited in a Bayesian manner by the AMP detector. Simulation results show that the one-bit magnitude measurements bring about high MIMO multiplexing and diversity gains, and decrease the transmission power. With the increase of the channel coherence time, more one-bit envelope chains are prone to be equipped, and one-bit magnitude-aided MIMO becomes more and more spectral-and-energy-efficient than the conventional MIMO.

Users require cloud providers to provide cloud services with suitable cost and acceptable reliability. They provide users the resources (e.g., bandwidth and processing) of virtual machines running on physical machines. A conventional virtual machine selection scheme adopts only a cloud provider that satisfies the acceptable reliability. The total cost of usage sometimes becomes unnecessarily high, since highly reliable cloud providers provide a high-cost service. This paper proposes a virtual machine selection scheme considering reliability for the cloud service. The proposed scheme satisfies the user's acceptable reliability using multiple cloud providers with a suitable cost and the acceptable reliability, while minimizing their total cost of usage. We formulate the virtual machine selection problem as a linear programming problem. Our simulation demonstrates that the proposed scheme reduces cost, compared to the conventional scheme.

Smart Open Shortest Path First (S-OSPF), which is an extended scheme of OSPF, was previously presented to avoid network congestion. However, it is difficult to achieve S-OSPF implementation, since the S-OSPF architecture is different from the conventional OSPF network architecture. A conventional network requires an autonomous distributed routing architecture, but S-OSPF partly requires a centralized routing architecture while keeping the feature of the conventional network. We employ software-defined networking (SDN) technology to support S-OSPF. Edge routers in the S-OSPF network must have both traffic distribution function, which is a feature of S-OSPF, and OSPF-based forwarding function, which is used in case that the edge router behaves as a transit router as is in the original OSPF network. A conflict occurs on the common forwarding table, which is accessed by SDN and OSPF, in the edge router when we try to utilize existing software modules. To solve this issue, this paper proposes an implementation method by introducing a hybrid router with virtualization technique. A hybrid router by the proposed method achieves the traffic distributing function of S-OSPF as well as the OSPF behavior while minimizing the modification of existing routing software modules. We develop a prototype of the hybrid router, confirm that the traffic distributing function of S-OSPF and OSPF functions work correctly, and observe that the congestion is avoided in our experimental network.

The recent development of network technologies that offer centralized control of explicit routes opens the door to the online optimization of explicit routing. For this kind of Traffic Engineering optimization, raising the calculation speeds by using multi-core processors with effective parallel algorithms is a key goal. This paper proposes an effective parallel algorithm for General purpose Programming on Graphic Processing Unit (GPGPU); its massively parallel style promises strong acceleration of calculation speed. The proposed algorithm parallelizes not only the search method of the Genetic Algorithm, but also its fitness functions, which calculate the network congestion ratio, so as to fully utilize the power of modern GPGPUs. Concurrently, each execution is designed for thread-block execution on the GPU with consideration of thread occupancy, local resources, and SIMT execution to maximize GPU performance. Evaluations show that the proposed algorithm offers, on average, a nine fold speedup compared to the conventional CPU approach.

Elastic Optical Networks (EONs) enable optical circuits to be assigned distinct numbers of spectrum slices. Individual circuits can then be assigned an optimal number of slices to best match their target transmission rates. A well-known drawback of EONs is spectrum fragmentation and its resulting uneven blocking probability, which circuit requests experience when the available spectrum slices in the fiber are insufficient or not contiguous. Capturing this spectrum fragmentation problem analytically is a challenging problem. Not surprisingly, most of the existing studies at this time mainly use simulation-based techniques to quantify blocking probability in EONs. In this paper, the authors present a Markov Chain (MC) model that attempts to characterize the fragmentation problem in a simplified scenario, i.e., only two types of circuit services are allowed over a single fiber link. Despite its limited scope, this initial analytical effort is able to accurately capture the non-monotonic behavior of the blocking probability in EONs for the first time.

In elastic optical networks (EONs), bandwidth fragmentation refers to the existence of non-aligned and noncontiguous subcarrier slots (unused) in the set of all subcarrier slots. Since wavelengths for a connection must be allocated on contiguous subcarrier slots, these non-aligned and noncontiguous available subcarrier slots could cause bandwidth blocking. This paper proposes a subcarrier-slot partition scheme for wavelength assignment in EONs that can yield more contiguous aligned available subcarrier slots, which reduces the bandwidth blocking probability. On this scheme, the total set of subcarrier slots is separated into several partitions and wavelengths are assigned to each partition based on the links utilized by particular connections. Numerical results from a simulation show that the proposed scheme with a suitable wavelength assignment policy outperforms the conventional scheme in terms of bandwidth blocking. We investigate the effect of different wavelength assignment policies in the proposed scheme. The results show that the first-last fit assignment policy gives lower bandwidth blocking probability than the first fit assignment policy.

This paper proposes a heuristic routing algorithm to design instantaneous recovery protection routes for all possible source destination pairs by provisioning network coding (NC) based 1+1 protection technique. We consider a static routing problem in networks where each node has the coding capability, and the exact traffic demand matrix is given. In the proposed heuristic algorithm a network with N nodes is divided into N scenarios, where each node is chosen as the common destination and k nodes among the remaining ones are the sources, where 2 ≤ k ≤N-1. By dividing the network into several scenarios with k sources and a common destination (k S D), all the possible source destination pairs, according to the given traffic matrix, are considered. It was reported that a mathematical programming approach to determine NC based 1+1 protection routes for any kSD scenario is an intractable problem for large k values. In the proposed heuristic algorithm we tackle this intractable problem by choosing either two or three sources out of k sources at a time according to the largest effective gain first policy, and then routing is assigned to the selected 2SD or 3SD scenario by using our developed mathematical models. The largest effective gain first policy ensures the best possible resource saving for each of the selected 2SD or 3SD scenario. We compare the total path costs of NC based 1+1 protection for all possible source destination pairs, obtained by our proposed heuristic algorithm, with that of the conventional 1+1 protection technique (without NC). Numerical results observes that almost 15% resource saving is achieved in our examined networks.

This paper proposes a survivable lightpath provisioning scheme that allows traffic splitting in multi-domain optical networks to minimize the cumulative cost of a set of paths. The proposed scheme, called two-phase lightpath provisioning' employs an integer linear programming (ILP) formulation based on hierarchical path computation with full-mesh topology abstraction. There are two phases in the scheme. The first phase solves the ILP problem on an inter-domain topology and then feeds the results as intra-domain requests. The second phase solves the ILP problem in each related domain. Finally, we concatenate all the intra-domain solutions along routing sequences. Three different protection strategies are considered with varying degrees of primary and backup route separation. Furthermore, to support various types of traffic demands, we investigate two cases in terms of the number of requested wavelengths. First, the number of requested wavelengths is less than link wavelength capacity. Second, the number of requested wavelengths is greater than link wavelength capacity. For the latter case, the proposed scheme allows traffic splitting among feasible primary and backup routes. The proposed scheme well supports the implementation of heuristic algorithms for lightpath provisioning since it can provide reference values, including upper and lower bounds, that are useful as benchmarks.

This paper proposes a heuristic routing algorithm to design routes for all possible source destination pairs by provisioning erasure correcting code based instantaneous recovery technique with optimal traffic splitting, which was addressed for a source destination pair in a prior work. We consider a static routing problem in networks having the coding capability. When the links in a network have finite capacities, assigning routing for all possible source destination pair by using this instantaneous recovery technique are mutually dependent, and this issue was not addressed. For the route designing purpose, one need to check routing for exponential number of traffic splitting number combinations. If the number of combinations to be considered, which equals the multiplication of each individual maximum possible traffic splitting numbers of all pairs considered, becomes extremely large obtaining a routing solution within a practical time is not possible. In order to achieve a routing solution within a practical time, the proposed heuristic algorithm gives highest priority to the pair either with the largest cost or with the largest resource saving effect. For all source destination pairs the total path costs of implementing erasure correcting code based instantaneous recovery technique, and conventional 1+1 protection technique are computed. Almost 20% resource saving w.r.to 1+1 protection is achieved in our examined networks.

The articles i nthis special issue focus on switching and routing applications for scalable and energy efficient networking.

This paper proposes a Preventive Start-time Optimization with no penalty (PSO-NP). The penalty being the generation of a higher than normal congestion ratio in non-failures scenario when the link weight set used in our network only targets the failure scenario. PSO-NP determines a suitable set of OSPF link weights at the start time that can handle any link failure scenario preventively while suppressing the penalty for the non-failure scenario. Previously, a preventive start time optimisation was presented to minimize the worst case congestion ratio in case of failure. That scheme unfortunately presents a non-negligible penalty when there is no link failure in the network because it only focuses on the failure scenario. In this paper we consider both the worst case failure scenario and the non-failure scenario.We suppress that penalty while enhancing the Preventive Start-Time scheme to counter failures. Simulation results show that PSO-NP achieves substantial congestion reduction for any failure case while eliminating the penalty in case of no failures in the network.

We demonstrate a distributed multicarrier reusable network (DMRN) for regional and metro areas, based on dense wavelength-division multiplexing (DWDM) transmission with reconfigurable optical add/drop multiplexers (ROADMs). To eliminate the multiple distributed laser-diodes (LDs) at each access node in conventional ROADM networks, optical carriers generated by a centralized multicarrier light source (MCLS) are distributed to the access nodes, and they are used for node-to-node data transmission. The ROADM employed at each access node is used not only to “add” and “drop” data, but also to “drop” optical carriers. Moreover, to improve the wavelength utilization efficiency of the carriers distributed by the MCLS in the network, we proposed a technique called optical carrier regeneration (OCR), whereby the distributed carriers can be reused in each access node. This technique has a simple scheme and enables us to reuse the carriers that were already utilized for data transmission between prior source and destination nodes. In this work, we numerically analyze the scalability of our proposed DMRN in terms of the number of nodes, the span length, and the cascadability of the OCR. Moreover, we conduct a DMRN experiment using 10.7 Gb/s × 4 channels DWDM transmission and compare the transmission performances for various span lengths, for the first time. The results show that the DMRN will be useful for wide-area metro networks with high transmission performances.

This paper proposes an even-split Smart-OSPF (S-OSPF) scheme to reduce network congestion more than the conventional non-split S-OSPF and to distribute traffic more easily than the conventional split S-OSPF. In split S-OSPF, source edge nodes distribute traffic unevenly to their neighbor nodes, but the implementation becomes involved to split traffic with different distribution. In non-split S-OSPF, source edge nodes transmit traffic to only one neighbor so that network congestion can be minimized, where non-split S-OSPF distributes traffic more simply than split S-OSPF. In the proposed scheme, source edge nodes transmit traffic evenly to selected neighbor nodes to minimize network congestion. The optimization problem to select a suitable set of neighbor nodes for even traffic distribution raised by the proposed scheme is formulated as an Integer Linear Programming (ILP) problem. The difficulty of solving the ILP problem in a practical time leads us to introduce a heuristic algorithm. The performances of our developed heuristic algorithm are evaluated via simulation developed in terms of network size. Numerical results show that even-split S-OSPF offers better routing performance than non-split S-OSPF for small-size networks and matches that of split S-OSPF for large-size networks.

This paper evaluates tradeoff between the partial disjoint path (PDP) and the k failure-disjoint path (k-FDP) algorithms for finding k disjoint paths that share reliable links, which have no failures. Adopting a suitable algorithm for each network scenario is able to save time for network operators or save cost for network providers. We provide examined data under sharing reliable links scenario for implementing survivability of networks. The PDP algorithm finds k disjoint paths with sharing reliable links. The FDP algorithm finds disjoint paths with minimum summation of two disjoint path costs. The k-FDP algorithm is extended from the original FDP algorithm. However, k-FDP takes longer time to find the disjoint paths than PDP for same cases. In this paper, computational time and summation of path costs are considered as performance indicators. The performance of both algorithms is examined using computer simulations with various network topologies. The simulation results show that the k-FDP algorithm takes a large amount of computational time to find k disjoint paths when the number of nodes increases. The PDP algorithm is able to reduce the computational time to find k disjoint paths in the networks by 99% compared to the k-FDP algorithm. However, the summation of path costs of PDP is higher when the number of required disjoint paths increases. Our comparative results assist network operators or planners to understand the tradeoff between PDP and k-FDP which in turn helps them to make an appropriate decision in network implementations.

Optimizing link weights in an OSPF network is a key traffic engineering problem to reduce the network congestion. Previous studies introduced three policies on link weight optimization, which are called Start-time Optimization (SO), Runtime Optimization (RO), and Preventive Start-time Optimization. All of them were used to be applied to a pipe mode, in which a traffic matrix, representing the traffic demand between each source and destination node pair, is exactly known. In practical, it is difficult for a network operator to measure or specify an exact traffic matrix. On the other hand, it is easy for a network operator to specify a hose model, in which only the total amount of traffic each node injects into the network and the total amount of traffic each node receives from the network, has to be known. This paper proposes a preventive start-time optimization scheme for the hose model. It employs a heuristic algorithm to determine an optimal set of link weights to minimize the worst-case congestion ratio for any single link failure. A straight-forward method to find an optimal set of link weights is to search the link weight space against all the possible traffic matrices and topologies created by link failure. Obviously, this approach needs a huge amount of computation time. The proposed scheme effectively selects the worst-case performance traffic matrix and tries to reduce the worst-case congestion ratio. Numerical results via simulations show that the proposed scheme effectively reduces the worst-case congestion ratio compared to SO.

This paper proposes a scheme for pump wavelengths selection in an optical packet switch (OPS) with parametric wavelength converters (PWCs), where the pump wavelengths are dynamically changed for all time slots and more than one PWC are allowed to convert a wavelength in a recursive manner. This scheme is called a dynamic pump-wavelength selection with recursive parametric wavelength conversion (DPS-R). A PWC, which has an advantage of multiple wavelength conversion, uses a pump wavelength that can be flexibly chosen to define which wavelengths can be converted from/to, called wavelength conversion pairs. The OPS allows each wavelength to be converted using combination of available conversion pairs from more than one PWC. A conventional scheme, pump wavelengths are statically preassigned, so that the conversion pairs are fixed for all time slots. Requests may remain since the pump wavelengths are not able to be reconfigured. The available conversion pairs may not support those requests. DPS-R is used to select the pump wavelength for each PWC to maximize the number of wavelength conversion pairs supported, in both recursive and non-recursive manners. Numerical results via simulation show that DPS-R outperforms the conventional scheme in term of packet loss rate.

In grid networks, there are different communication pairs between senders and receivers, where they communicate simultaneously. These different simultaneous communications are called multiple-pair communications. In multiple-pair communications, an identical link that is shared on paths of different communications may exist. The link is called a common link. A controller of multiple-pair communications needs to know the link available bandwidth of the common link, which may limits the available bandwidth in simultaneous multiple-pair communications, for scheduled communications. An objective of scheduling is to complete all required communications as quickly as possible when each traffic demand is given. A conventional scheme is not able to estimate the link available bandwidth of a common link. This paper proposes a scheme for estimating an available bandwidth in simultaneous multiple-pair communications, which is called a simultaneous available bandwidth. To achieve this, the proposed scheme estimates the link available bandwidth of a common link by synchronizing packet streams at the common link if any common link exists and the bandwidth limits the simultaneous available bandwidth. The proposed scheme employs the metrics that are used to estimate the path available bandwidth of a single-pair communication in the conventional scheme, to synchronize packet streams. This paper formulates an optimal adjustment width of transmission time of a packet stream to synchronize packet streams when a target synchronization ratio is given.

This paper presents an optimal routing strategy based on the Hose model with bounds of Link Traffic (HLT), which we introduce. HLT is specified by the total traffic passing through each link in addition to the traffic bounds described in the hose model. The pipe model, which is specified by the exact traffic matrix, provides the best routing performance, but the traffic matrix is difficult to measure and predict accurately. While the hose model employs just the total outgoing/incoming traffic from/to each node, it offers lower routing performance than the pipe model, due to insufficient traffic information. The Hose model with bounds of Source-Destination Traffic (HSDT), where the upper and lower bounds of traffic demands for source-destination pairs are added as constraints, is a construction that lies between the pipe and hose models, but determining additional bounds is not easy for the network operators to specify. HLT, which lightens the difficulty of the pipe model, but narrows the range of traffic conditions specified by the hose model, offers better routing performance than the hose model. In addition, the HLT model resolves the difficulty of the HSDT model with regard to determining appropriate additional bounds. An optimal-routing formulation extended from the pipe model to the HLT model can not be solved as a regular linear programming (LP) problem. Our solution, the introduction of a duality theorem, turns this problem into an LP formulation that can be easily solved. Numerical results via simulations show that HLT offers 20-35% lower network congestion ratios than the hose model. In addition, the congestion ratios of the pipe and HLT models differ by less than 0.1.

Out-of-sequence is a problem faced by multi-stage buffered Clos-network switches. This paper proposes two buffered three-stage Clos-network packet switches that service packets in sequence and provide high switching performance. The proposed switches require short configuration times as compared to existing bufferless or partially buffered Clos-network switches. The proposed switches use time stamps assigned at the input modules to identify the order of packets in the switch. The switches use time-stamp monitoring mechanisms either at the input modules in a switch called the MMM-IM switch, or at the output modules in a switch called the MMM-OM switch to keep packets in sequence. Synchronization among different switch modules is not required in the proposed switches. The switching performance study presented in this paper shows that in-sequence monitoring at the IM provides higher performance and larger scalability than in-sequence monitoring at the output. Furthermore, the throughput of the MMM-IM switch is comparable to that of a switch that may service packets out of sequence.

A key traffic engineering problem in the Open Shortest Path First (OSPF)-based network is the determination of optimal link weights. From the network operators' point of view, there are two approaches to determining a set of link weights: Start-time Optimization (SO) and Run-time Optimization (RO). We previously presented a Preventive Start-time Optimization (PSO) scheme that determines an appropriate set of link weights at start time. It can counter both unexpected network congestion and network instability and thus overcomes the drawbacks of SO and RO, respectively. The previous work adopts a preventive start-time optimization algorithm with limited candidates, named PSO-L (PSO for Limited candidates). Although PSO-L relaxes the worst-case congestion, it does not confirm the optimal worst-case performance. To pursue this optimality, this paper proposes a preventive start-time optimization algorithm with a wide range of candidates, named PSO-W (PSO for Wide-range candidates). PSO-W upgrades the objective function of SO that determines the set of link weights at start time by considering all possible single link failures; its goal is to minimize the worst-case congestion. Numerical results via simulations show that PSO-W effectively relaxes the worst-case network congestion compared to SO, while it avoids the network instability caused by the run-time changes of link weights caused by RO. At the same time, PSO-W yields performance superior to that of PSO-L.

This paper proposes a scheme to increase possible patterns of wavelength-conversion for an optical packet switch (OPS) with parametric wavelength converters (PWCs) by obtaining additional converted wavelengths using the existing resources. It is called a recursive parametric wavelength conversion (RPWC) scheme. A PWC uses a pump wavelength to define the original and converted wavelengths, called wavelength conversion pairs. None of conversion pairs from any PWCs can sometime support the original and available converted wavelength, although some wavelengths at the requested output fiber are available. Since the original wavelength is not converted, some packet losses may occur. Several conversion pairs are wasteful since they are not utilized. In RPWC scheme, unused conversion pairs are used to create additional conversion pairs. The OPS allows each wavelength to be converted using combination of unused conversion pairs using more than one PWCs, instead of using only a single PWC as in a conventional scheme. Numerical results via simulation show that the RPWC scheme achieves lower packet loss rate than the conventional scheme. To show the feasibility of the RPWC scheme, we develop a prototype of an optical switch with RPWC and demonstrate it in experiment.

This paper proposes a scheme to find k disjoint paths in multi-cost networks. This scheme, called the k-penalty scheme with initial arc cost matrix (KPI), penalizes the use of conflicting arcs found in previously set paths and increases the costs of these arcs in accordance with the initially given arc cost matrix. Simulations show that the KPI scheme is able to find k disjoint paths faster than the conventional scheme that uses the incrementally updated auxiliary arc cost matrix to increases the cost of conflicting arcs. Moreover, the KPI scheme yields k disjoint paths with lower total cost than the conventional scheme.

Smart OSPF (S-OSPF), a load balancing, shortest-path-based routing scheme, was introduced to improve the routing performances of legacy networks running OSPF with known traffic demands. S-OSPF distributes traffic from a source node to neighbor nodes, and, after reaching the neighbor nodes, traffic is routed according to the OSPF protocol. However, in practice, exact traffic demands are difficult to obtain, and most routers will not be able to handle the complexity of determining and implementing uneven traffic distributions with any form of precision. This paper investigates non-split S-OSPF with the hose model for the first time; its goal is to minimize the worst-case network congestion ratio. In this model, traffic from a source node to a destination node is not split over multiple routes, in other words, it goes via only one neighbor node to the destination node. The routing decision problem with the hose model is formulated as an integer linear programming (ILP) problem. Since it is difficult to solve the ILP problem in practical time, this paper proposes a heuristic algorithm. In the routing decision process, the proposed algorithm gives the highest priority to the node pair that has the highest product of ingress and egress traffic, and enables a source node to select the neighbor node that minimizes the maximum link utilization over all links for the worst case traffic condition specified by the hose model. We compare non-split S-OSPF to split S-OSPF and classical shortest path routing (SPR). Numerical results show that the non-split S-OSPF scheme improves routing performance versus classical SPR and is comparable to the split S-OSPF scheme for larger networks.

Streaming services such as voice and video based on IP technologies have been deployed widely throughout the world. The Asymmetric Digital Subscriber Line (ADSL) is one of broadband access technologies used to offer these services. Network operators and/or integrators must estimate ADSL link rates from the Network Operation Centers (NOC) which are far away from the customer's site. They must also use a simple measuring scheme so that existing network node facilities are not unduly burdened. Unfortunately, there is no practical scheme that can satisfy these requirements. This letter proposes a practical scheme that can estimate ADSL link rates. The proposed scheme allows us to estimate ADSL link rates from measurements made at the NOC using existing communications protocols and network node facilities; it imposes no heavy traffic overhead. The proposed scheme consists of two major steps. The first step is to collect measured data of round trip times (RTT) for both long and short packets to find their minimum values of RTTs, i.e. those do not include queuing delays. The RTT measurements are simply conducted by sending Internet Control Message Protocol (ICMP) echo request messages. The second step is to estimate the ADSL down- and up-link rates by using the difference in RTT between long and short packets and the experimentally-obtained correlated relationships between ADSL down- and up-link rates. RTTs are experimentally measured for an IP network, and it is shown that the down- and up-link rates can be obtained in a simple manner.

This paper proposes an optical packet switch with parametric wavelength converters (PWCs), where the pump wavelengths are dynamically changed in every time slot. It is called the dynamic pump wavelength selection (DPS) switch. A PWC, which performs multiple wavelength conversion, uses a pump wavelength that can be flexibly chosen to define which wavelengths can be converted. To enhance switch performance, the DPS switch employs a matching scheme, which sets connections between input and output ports, in combination with dynamic pump wavelength selection; a conventional switch, on the other hand, performs matching with a given set of pump wavelengths that are configured in a static manner. The dynamic pump wavelength selection is used to select the pump wavelength for each PWC to maximize the number of wavelength conversion pairs supported. Numerical results from a simulation show that the DPS switch provides lower blocking rates than the conventional switch with pump wavelengths assigned statically, under both uniform and non-uniform traffic.

One-Way Delay Variation (OWDV) has become increasingly of interest to evaluate network state and service quality, especially for real-time and streaming services such as VoIP and video. Measurement of these parameters needs to be performed with the layout infrastructure. Many schemes for OWD measurements require clock synchronization at the source and destination through Global-Positioning System (GPS) or the Network Time Protocol (NTP). In clock-synchronized approaches, the accuracy of the measurement of OWDV depends on the achieved accuracy of clock synchronization. GPS provides high-accuracy clock synchronization. However, the deployment of GPS on legacy network equipment might be slow and costly. This paper proposes a method for measuring OWDV without recurring to clock synchronization. However, clock skew may affect the measurement of OWDV. The proposed approach is based on the measurement of Inter-Packet Delay (IPD) and Accumulated OWDV (AOWDV). This paper shows the performance of the proposed scheme via simulation and through experimentation in a VoIP network. The presented simulation and experimental results indicate that clock skew can be efficiently measured and removed and that OWDV can be measured without requiring clock synchronization.

We propose a new efficient singlecast / multicast method for active optical access network using PLZT 10 nsec high-speed optical switches. The Active Optical Network, called ActiON, has been proposed using slot type switched optical network. Compared with Passive Optical Network (PON), ActiON can quadruplicate the number of subscribers (128 users) per OLT and double the maximum transmission distance (40 km) between OLT and ONUs. However, ActiON uses slot based switching method, so it is difficult to realize the multicast delivery. In this paper, we propose the efficient singlecast / multicast method for ActiON by using PLZT optical switch elements which are controlled as “distribution mode” like an optical splitter by applying mid-control voltage. In addition, a new efficient multicast slot allocation method is proposed and formulated as a linear programming problem. This formula develops the maximum number of users which is able to be connected by multicast and the minimum number of slots is used for multicast users.

In an optical packet switch (OPS), input fibers carry multiple wavelengths, which carry packets to one or more output fibers. As several wavelengths from different inputs could be destined to the same output fiber, one wavelength can be connected and the others remain disconnected, losing the carried packets. Because of the multiple wavelengths available at an output fiber, wavelength conversion in the OPS of the unconnected wavelengths into those available can increase the number of connections. A parametric wavelength converter (PWC) provides multi-channel wavelength conversion where wavelengths can be converted to another. A PWC uses a pump wavelength that can be flexibly chosen to define which wavelengths can be converted, defining the so-called wavelength conversion pairs. However, it is unknown which set of pump wavelengths, and therefore the set of connection pairs, should be selected to improve the OPS performance while minimizing the number of PWCs in the OPS. Therefore, this paper proposes a pump wavelength selection policy for an OPS that uses different pump wavelengths, one for each PWC, within an arbitrarily selected interval. This policy is called variety rich (VR) policy. This paper also introduces a non-wavelength blocking OPS (NWB-OPS) to make full use of PWCs. The switch performance is evaluated through computer simulation. The results show that the proposed policy with different pump wavelengths achieves the highest performance when compared to another of similar complexity. Furthermore, the performance study shows that small sizes of the interval to select a pump wavelength are more beneficial than larger ones.

This paper presents and experimentally demonstrates a multi-carrier distributed wavelength-division-multiplexing (WDM) ring network based on reconfigurable optical "drop-add-drop" multiplexers for regional and metro network applications. In the "drop-add-drop" network, optical carriers generated by a centralized multi-carrier light source (MCLS) are "dropped" at the source nodes and used for uplink transmission. Data are "added" to the network by external modulation of one or more carriers. Data are then "dropped" at the destination nodes. The reconfigurable optical add/drop multiplexer (ROADM) at each access node is not only used to "add" and "drop" data, but also to "drop" carriers, which eliminates the many distributed laser-diodes used in the conventional network. In this work, we successfully demonstrate, for the first time, a "drop-add-drop" network experiment offering 10 Gbit/s WDM transmission. Moreover, to dramatically improve the utilization efficiency of the carrier wavelengths distributed by the MCLS in the "drop-add-drop" network, we introduce the carrier wavelength reuse technique which sets the carrier extraction circuits in each access node. This technique enables us to reuse the carrier wavelengths that were already utilized for data transmission between prior source and destination nodes. To evaluate the effect of carrier wavelength reuse, we compare the blocking probabilities of the "drop-add-drop" networks with and without carrier wavelength reuse. The results show that wavelength reuse dramatically reduced the blocking probability. In addition, we numerically analyze the advantages of the "drop-add-drop" network over the conventional ROADM network in terms of network cost and power consumption.

The fine two-phase routing (F-TPR) scheme, an IP finely-distributed load-balanced routing scheme based on two-phase routing over shortest paths, was previously presented to improve routing performances. F-TPR distributes traffic from a source node to intermediate nodes simply by using IP tunnels. F-TPR provides comparable routing performance to the sophisticated traffic engineering (TE) scheme of Multi-Protocol Label Switching (MPLS-TE). However, in practice, most routers will not be able to handle the complexity of determining and implementing uneven traffic distributions with any form of precision. This paper investigates non-split F-TPR, where traffic from a source node to a destination node is not split over multiple routes, in other words, it goes via only one intermediate node to the destination node. The problem solved by non-split F-TPR is formulated as an integer linear programming (ILP) problem. Since it is difficult to solve the ILP problem within a practical time, this paper introduces two heuristic algorithms against the ILP problem. We compare non-split F-TPR against split F-TPR and MPLS-TE. Numerical results show that non-split F-TPR matches the routing performance of F-TPR and MPLS-TE with an error of 1%, when network size is enough large.

This paper proposes a scalable network emulator architecture to support IP optical network management. The network emulator uses the same router interfaces to communicate with the IP optical TE server as the actual IP optical network, and behaves as an actual IP optical network between the interfaces. The network emulator mainly consists of databases and three modules: interface module, resource simulator module, and traffic generator module. To make the network emulator scalable in terms of network size, we employ TCP/IP socket communications between the modules. The proposed network emulator has the benefit that its implementation is not strongly dependent on hardware limitations. We develop a prototype of the network emulator based on the proposed architecture. A virtual machine (VM) technology is employed to reduce the hardware amounts required for this implementation. Our design and experiments show that the proposed architecture is effective. Thanks to the scalability and flexibility of the proposed architecture, it is expected that network-size can be easily scaled up.

This paper proposes a simple shortest-path-based load-balanced Internet-Protocol (IP) routing scheme based on the hose model. The proposed scheme is an extension of the Smart-OSPF (S-OSPF) scheme. The proposed scheme, the same as S-OSPF, splits traffic demand only at source edge nodes and transmits the traffic along the shortest path routes. In S-OSPF, the split ratios are determined for each source-destination edge node pair by assuming that the traffic demand between all source-destination edge node pairs are known, in other words, the exact traffic matrix is completely given. This, however, is difficult to measure and predict accurately because of the measurement costs and rapid traffic fluctuations. On the other hand, in the proposed scheme, we assume the use of the hose model; in this model, only the total amount of traffic that a node injects into the network and the total amount of traffic it receives from the network are known. This simplicity makes it easy for network operators to apply the hose model for IP routing. This is because measuring just the total amount of traffic is less expensive than measuring the traffic demands between all source-destination edge node pairs. In addition, the aggregated traffic exhibits less fluctuation and is easier to predict than the traffic demand between each source-destination pair. Any extension of the Linear Programming (LP) formulation that optimizes S-OSPF to suit the hose model cannot be solved as a simple LP problem, because the traffic matrix is not known. By introducing a duality theorem, we successfully formulate our problem as an LP formulation that can be easily solved yielding the desired split ratios. Numerical results show that the proposed scheme dramatically reduces the network congestion ratio compared to the classical shortest path routing scheme and it provides performance close to that provided by the sophisticated traffic-engineering (TE) scheme of Multi-Protocol Label Switching (MPLS)-TE.

This paper proposes an optical broadcast-and-select network architecture with centralized multi-carrier light source (C-MCLS). A large number of optical carriers/wavelengths generated by C-MCLS are distributed to all edge nodes (ENs), which select and modulate wavelengths to realize transmission. To utilize wavelength resources efficiently, we introduce a framework of wavelength allocation and selection (WAS). Wavelength allocation is performed at a wavelength control server, while wavelength selection is done at each EN according to wavelength allocation results. Both static and dynamic schemes are adopted for WAS and their implementations are shown. By using fixed or tunable band pass filter and periodical arrayed waveguide grating demultiplexer, wavelengths are selected and utilized by ENs in a static or dynamic manner. We evaluate network cost and performance of the proposed network. Cost analysis and numerical results show that it offers greatly reduced cost compared to the conventional one when the number of required access wavelengths at EN becomes large. We delineate its applicable areas through cost comparisons. Blocking probabilities of static and dynamic schemes are analyzed to evaluate network performance. Numerical results show that by choosing appropriate design parameters, the dynamic scheme offers about 25% increase in admissible offered load under the specified blocking probability, compared to the static scheme. This indicates that the dynamic scheme makes the proposed network more robust against traffic fluctuations.

This paper proposes a simple shortest-path-based load-balanced IP routing scheme for the hose model. The proposed scheme is an extension of the Smart-OSPF scheme. The proposed scheme, the same as S-OSPF, splits traffic demand only at source edge nodes and transmits the traffic along the shortest path routes. In S-OSPF, the split ratios are determined for each source-destination edge node pair by assuming that the traffic demand between all source-destination edge node pairs are known, in other words, the exact traffic matrix is completely given. On the other hand, in the proposed scheme, we assume the use of the hose model; in this model, only the total amount of traffic that a node injects into the network and the total amount of traffic it receives from the network are known. Any extension of the linear programming (LP) formulation to suit the hose model cannot be solved as a simple LP problem, because the traffic matrix is not known. By introducing a duality theorem, we successfully formulate our problem as an LP formulation that can be easily solved yielding the desired split ratios. Numerical results show that the proposed scheme dramatically reduces the network congestion ratio compared to the classical shortest path routing scheme and it provides performance close to that provided by the sophisticated traffic-engineering (TE) scheme of multi-protocol label switching (MPLS)-TE.

The performance of matching schemes for input- queued (IQ) packet switches is mainly defined by the selection policy adopted. This policy can be aimed to produce a large weight sum for matched input-output pairs, where each input- output pair is assigned a weight, or to produce a large match size in the number of matched pairs, giving place to maximum weight matching or maximum size matching, respectively. However, schedulers can only provide a single match in function of the selection (of candidate ports) policy adopted and of the backlogged traffic at the input queues. A parallel match evaluator was recently proposed to provide not one but several match options at the same time. This approach evaluates several predefined and fixed matches and picks the match with the largest size. However, the fixed permutations of the evaluated matches may produce low performance under traffic with nonuniform distributions because of the limited number of choices. This paper proposes to make the parallel match evaluator configurable and two schemes to provide diverse and changeable matches such that the matches (and therefore, the evaluator) become adaptable to the traffic pattern. The proposed schemes were tested under uniform and nonuniform traffic patterns and the results show that these schemes provide high performance, even when scheduling is performed between periods of multiple time slots, or framed intervals. The proposed approach can be used for configuring slow micro-electro-mechanical (MEM) optical switch fabrics.

This paper proposes a real-time allocation scheme for photonic networks that use wavelength division multiplexing (WDM) and optical time division multiplexing (OTDM) technologies in the backbone and ring regional networks, respectively. A frame that is used for transferring data and control information in a ring regional network takes only one round to complete data allocation processing, instead of two rounds as required by the prior reservation scheme, so that data can be transmitted immediately. Our challenge is to provide max-min fair share in terms of throughput with just one round. If no free space is left on the frame, the proposed scheme allows a group (some) of the newly requested data to replace some of the already allocated data to provide max-min fair share, in terms of throughput. Data replacement and de-fragmentation are processed in the optical domain. Simulations show that the proposed scheme maintains max-min fair share even in unbalanced traffic scenarios. The complexity of de-fragmentation depends on the number of delay lines needed to regenerate the original data groups. The maximum number of delay lines is determined.

This paper proposes an IP finely-distributed load- balanced routing scheme based on two-phase routing over shortest paths, where the traffic matrix is given. It is called the fine two-phase routing (F-TPR) scheme. In F-TPR, traffic is distributed from a source node to intermediate nodes more finely, compared to the original TPR. F-TPR determines the distribution ratios to intermediate nodes for each source-destination node pair independently. To determine an optimum set of the distribution ratios, a linear programming (LP) formulation is derived. We compare the F-TPR scheme against the TPR scheme and the sophisticated traffic engineering (TE) scheme of multi-protocol label switching (MPLS-TE). Numerical results show that F-TPR greatly reduces the network congestion ratio compared to TPR. In addition, F-TPR provides almost the same network congestion ratio as that of MPLS-TE, the difference is surprisingly less than 0.1% for various experimented network topologies.

The throughput of a packet switch is a major switch property, and therefore, of major interest to analyze it. An approximation of the throughput of a staged random selection algorithm with a single iteration under uniform for a three-stage Clos-network packet switch, also called a Space- Space-Space (S3) Clos-network packet switch, has been recently presented. However, the difference between this approximation and the actual throughput of the staged random selection algorithm is significant. To address this issue, this paper presents a theoretical throughput analysis of the staged random selection algorithm with a single iteration for a S3Clos-network switch and show that the throughput is higher than that estimated by the existing approximation. Second, the paper extends the analysis to calculate the throughput of the staged random selection algorithm with multiple iterations by considering the analysis of the parallel iterative matching scheme, which is a random-based matching scheme for single-stage switches. The introduced derivation carefully considers the behavior of the selection algorithm at the switching modules in all three stages of the switch. The probability that a request reaches the third-stage modules is affected by the matching results at the second-stage modules. Numerical evaluations of the analytical formulas are performed. The results show that the staged random selection algorithm with multiple iterations for a S3Clos-network switch without internal expansion can achieve 100% throughput under uniform traffic.

This paper presents an optimization of IP load-balanced routing for the hose model. We present an IP load-balanced routing scheme based on the two-phase routing over shortest paths. It is called a fine two-phase routing (F-TPR) scheme. In F-TPR, traffic is distributed from a source node to intermediate nodes more finely, compared to the original TPR. F-TPR introduces the distribution ratio to node m that is determined for each source-destination pair of (p, q), kmpq. To determine an optimum set of kmpq, an linear programming (LP) formulation is first derived. However, the formulation is difficult to solve as a simple LP problem. This is because each element of the traffic matrix is not determined because of the hose model and there are too many possible parameters for us to consider. By introducing a duality theorem , we successfully formulate our problem a quadratic constraint programming (QCP) formulation that can be solved to determine the split ratios by using a mathematical programming solver. We compare F-TPR with TPR and the multi-protocol label switching (MPLS)-traffic engineering (TE). Numerical results show that F-TPR reduces the network congestion ratio compared to TPR. Numerical results show that F-TPR greatly reduces the network congestion ratio compared to TPR , and provides the network congestion ratio close to that of MPLS-TE within the difference of 6%.

Obtaining current traffic matrices is essential to traffic engineering (TE) methods. Because it is difficult to monitor traffic matrices, several methods for estimating them from link loads have been proposed. The models used in these methods, however, are incorrect for some real networks. Thus, methods improving the accuracy of estimation by changing routes also have been proposed. However, existing methods for estimating the traffic matrix by changing routes, however, can only capture long-term variations and cannot obtain current traffic matrices accurately. In this paper, we propose a method for estimating current traffic matrices by using route changes introduced by a TE method. In this method, we first estimate the long-term variations of traffic by using the link loads monitored the last M times. Then, we adjust the estimated long-term variations so as to fit the current link loads. In addition, when the traffic variation trends change and the estimated long-term variations cannot match the current traffic, our method detects mismatches. Then, so as to capture the current traffic variations, the method re-estimates the long-term variations after removing information about the end-to-end traffic causing the mismatches. For this paper, we evaluated our method through simulation. The results show that our method can estimate current traffic matrices accurately even when some end-to-end traffic changes suddenly.

This paper proposes a new scheme for diverse path setup in multi-domain optical networks, and presents its applicability along with other existing schemes. Ability to setup diverse paths is an important feature to improve resiliency. Currently, the Internet Engineering Task Force (IETF) is standardizing tools for automating optical path provisioning across multiple domains, such as Generalized Multi-Protocol Label Switching (GMPLS) signaling to instantiate optical paths, and Path Computation Element (PCE) to perform path computation. However, details on diverse path setup schemes require further analysis. There are several existing schemes for diverse path setup, but they have deficiencies. An enhanced scheme is proposed to overcome such deficiencies. Applicability of the proposed scheme is presented along with existing schemes by considering various aspects. This includes quantifying the ability to find diverse paths by simulations. Furthermore, this paper presents challenges for further study, including specific issues in multi-domain transparent optical networks and inter-carrier considerations.

We have been developing a network visualization technique: a network architecture technology to overlay IP networks over an optical backbone network to improve flexibility and resiliency. Virtual network topology created in the optical backbone network is used as a substrate for a IP network. In order to improve flexibility and resiliency, we need to implement carrier-grade mechanisms for operation, administration, and maintenance. For the IP network, the following management and controls are addressed: traffic measurement, route trace, topology and route optimization, and failure analysis. For the optical backbone network, the impact of the GMPLS control plane is addressed. We also discuss the integrated operation and management of the IP and optical backbone networks.

This paper proposes a network design method based on adaptive selection of "facility-adding" and "path-routing" policies under the condition that traffic keeps increasing. In a network where a path is provided as a service, when a new path demand is generated and if it is impossible to accommodate the path along the shortest route with only existing facilities (links and nodes, etc.), there are two policies to accommodate this new path demand. One is a facility-adding policy, which accommodates the path along the shortest route by adding facilities. The other is a path-routing policy, which finds the detour route that meets the bandwidth demand of the path and accommodates it along this detour route without adding facilities. The proposed network design method adaptively selects which policy to be applied per path according to the holding time of the path. Therefore, it is expected that the total facility cost are reduced compared to the conventional network design method, which uses only either one of two policies. Simulation results show that the proposed method contributes to the total facility cost reduction for an arbitrary design period and achieves 10% total facility cost reduction compared to the conventional method.

In a service provider (SP) network, routes for external destinations are distributed on iBGP sessions. This traditionally required the establishment of a full-mesh of iBGP sessions in the network. A common practice is now to make use of route reflectors (RR). Such a practice is more scalable in the number of iBGP sessions to be configured in a SP network. However, it has been shown that RRs have a negative impact on the diversity of routes available in the network. This is an important issue as routers may not be able to quickly use an alternate route in case of a route failure. In this paper we tackle the problem of route diversity in a service provider network composed of RRs. We propose an algorithm to design iBGP session topologies with improved route diversity. We rely on an initial route reflection topology. Our algorithm proposes the addition of a few iBGP sessions to some border routers of the domain. These border routers receive a large number of external routes for which routers lack diversity. We show by means of simulations that our algorithm meets its goals. In the resulting topologies, each BGP router knows at least two different ways to reach distant destinations. This is ensured as long as a prefix advertisement is received at different nodes at the border of the AS. Secondly, we observe that the number of iBGP sessions required to achieve this goal is significantly below the number of sessions required in the case of a full-mesh. Finally, the remaining lack of route diversity after the use of our design algorithm indicates that new external peering sessions should be established. In this case, our algorithm shows that diversity cannot be reached for some prefixes independently of the iBGP topology, with the current external peering sessions.

Overlay networks achieve new functionality and enhance network performance by allowing routing to be controlled at the application layer. However, these approaches result in degradations of underlying networks due to the selfish behavior of overlay networks. In this paper, we investigate the stability of virtual network topology (VNT) control under the overlay networks that perform dynamic routing updates. We reveal that the dynamics of routing on overlay networks causes a high fluctuation in the traffic demand matrix, which leads to significant instability of VNT control. To overcome the instability induced by the overlay routing, we introduce hysteresis to the VNT control. Simulation results indicate that the hysteresis mechanism improves the network stability, but cannot always improve the network performance. We therefore extend the hysteresis mechanism and show that the proposed method improves both the network stability and the performance when the amount of traffic for overlay network is not large.

This paper presents multi-layer IP optical traffic engineering experimental results using our developed IP optical Traffic Engineering (TE) Server, which performs traffic control in corporation with IP routes and optical cross-connects (OXCs) in Multi-Protocol Label Switching (MPLS) and Generalized MPLS (GMPLS) networks. Multi-layer TE can optimize network resource utilization considering all layers, rather than performing optimization independently for each layer. IP routers and OXCs are managed and operated by the IP optical TE server considering the network resources of both layers. The IP optical TE server computes path routers across different layers and controls them upon request from users and operators. The IP optical TE server dynamically reconfigures an IP network topology that consists of several optical paths in response to traffic demand fluctuations and network failures. The IP optical TE server dynamically reconfigures an IP network topology that consists of several optical paths in response to traffic demand fluctuations and network failures. We successfully performed experiments multi-layer TE.

In this paper, we present a practical VNT (virtual network topology) reconfiguration method for large-scale IP and optical networks with traffic matrix estimation considerations. We newly introduce a partial VNT reconfiguration algorithm with multiple transition stages. By dividing the whole VNT transition sequence into multiple transitions, estimation errors are calibrated at each stage by using network state information of prior stages. Because estimation errors are mainly due to the fewer information in the estimated traffic matrix calculation, our approach tries to increase the constraint conditions for traffic matrix estimation by introducing partial reconfiguration, and to relax the impact of estimation errors by limiting the number of optical-paths reconfigured at each stage. We also investigate the effectiveness of our proposal through simulations and clarify the robustness against estimation errors by using partial reconfiguration.

Input-buffered switches have been widely considered for implementing feasible packet switches. However, their matching process may not be time-efficient for switches with high-speed ports. Buffered crossbars (BXs) are an alternative to relax timing for packet switches with high-speed ports and to provide high-performance switching. BX switches were originally considered expensive, as the memory amount required in the crosspoints (XPs) is proportional to the square of the number of ports (O(N/sup 2/)). This limitation is now less stringent with the advances on chip-fabrication techniques, and when considering small crosspoint (XP) buffer sizes. In this paper, we study a combined input-crosspoint buffered packet switch, named CIXB, with virtual output queues (VOQs) at the inputs, and arbitration based on round-robin selection. We show that the CIXB switch achieves 100% throughput under uniform traffic, and high performance under nonuniform traffic, using one-cell XP buffer size and no speedup.

The scalability of Clos-network switches makes them an alternative to single-stages switches for implementing large-size packet switches. This paper introduces a cell dispatching scheme, called Maximum Weight Matching Dispatching (MWMD) scheme, for buffered Clos-network switches. The MWMD scheme is based on a maximum weight matching algorithm for input-buffered switches. This paper shows that, with request queues in the buffered Clos-network architecture, the MWMD scheme is able to achieve a 100% throughput for independent admissible traffic, without allocating any buffers in the second stage and without expanding the internal bandwidth. As a practical scheme, a maximal oldest-cell-first matching dispatching (MOMD) scheme is also introduced. MOMD shows that using a finite number of iterations in the dispatching scheme, the throughout under unbalanced traffic pattern can be high.

The paper proposes the hierarchical cloud-router network (HCRN) to solve the problem of overcoming the scalability limit in a multi-layer generalized multi-protocol label switching (GMPLS) network. We define a group of nodes as a virtual node, called a cloud-router (CR). A CR consists of some number of nodes or lower-level CRs. A CR is modeled as a multiple switching capability (SC) node when it includes more than one kind of SC, such as fiber SC, lambda SC, time-division multiplexing (TDM) SC, packet SC, even if there are no actual multiple-SC nodes in the CR. The CR advertises its abstracted CR internal structure which is abstracted link state information inside the CR. A large-scale, multi-layer network can then achieve scalability by advertising the CR internal structure throughout the whole network. In this scheme, the ends of a link connecting two CRs are defined as interfaces of the CRs. We adopt the CR internal cost scheme between a CR's interfaces to abstract the network. This CR internal cost is advertised outside the CRs via the interfaces. Our performance evaluation has shown that HCRN can operate a larger number of nodes than a normal GMPLS network. It can also bear more frequent network topology changes than a normal GMPLS network.

This paper proposes a distributed virtual network topology (VNT) reconfiguration method for Internet Protocol over a wavelength-division-multiplexing network under dynamic traffic demand. We have developed a simple heuristic algorithm for calculating the VNT for distributed control. A generalized multiprotocol label switching (GMPLS)-based routing protocol has been developed. The VNT is quickly reconfigured by setting up and/or tearing down lightpaths using a GMPLS signaling protocol. Traffic demand is measured at the ingress node and advertised by the extended GMPLS routing protocol. Performance of the proposed method is investigated using variable traffic model.

In high-speed and high-capacity packet switches, system reliability is critical to avoid loss of huge amounts of information and retransmission of traffic. We propose a series of concurrent fault-detection mechanisms for a multiple-plane crossbar-based packet switch. Our switch model, called the m+z model, has m active planes and z spare planes. This switch has distributed arbiters on each plane. The spare planes, used for substitution of faulty active ones, are also used in the fault-detection mechanism, thus providing fault detection and fault location for all switching planes. Our detection schemes are able to detect a single fault quickly without increasing transmission overhead. The proposed schemes can be used for switches with different numbers of active planes and a small number of spare planes.

The paper describes multi-layer traffic engineering and signaling technologies in photonic-GMPLS-router networks. Multi-layer traffic engineering, which yields the dynamic cooperation of IP and photonic layers for providing IP services cost-effectively, is described. To realize multi-layer traffic engineering, we propose an OSPF extension that advertises both the number of total wavelengths and the number of unused wavelengths and an RSVP-TE extension that minimizes the number of wavelength conversions needed. In addition, the paper presents a heuristic-based multi-layer topology design scheme that uses IP traffic measurements in a generalized multiprotocol label switch (GMPLS). Our design scheme yields the optical label switch path (OLSP) network topology, i.e. OLSP placement, that minimizes network cost, in response to fluctuations in IP traffic demand. In other words, the OLSP network topology is dynamically reconfigured to match IP traffic demand. Networks are reconfigured by the proposed scheme so as to utilize the network resources in the most cost-effective manner.

This paper proposes a pipeline-based concurrent round-robin dispatching scheme, called PCRRD, for Clos-network switches. Our previously proposed concurrent round-robin dispatching (CRRD) scheme provides 100% throughput under uniform traffic by using simple round-robin arbiters, but it has the strict timing constraint that the dispatching scheduling has to be completed within one cell time slot. This is a bottleneck in building high-performance switching systems. To relax the strict timing constraint of CRRD, we propose to use more than one scheduler engine, up to P, so called subschedulers. Each subscheduler is allowed to take more than one time slot for dispatching. Every time slot, one out of P subschedulers provides the dispatching result. The subschedulers adopt our original CRRD algorithm. We show that PCRRD preserves 100% throughput under uniform traffic of our original CRRD algorithm, while ensuring the cell-sequence order. Since the constraint of the scheduling timing is dramatically relaxed, it is suitable for high-performance switching systems even when the switch size increases and port speed is high (e.g., 40 Gbit/s).

A Clos-network switch architecture is attractive because of its scalability. Previously proposed implementable dispatching schemes from the first stage to the second stage, such as random dispatching (RD), are not able to achieve high throughput unless the internal bandwidth is expanded. This paper presents two round-robin-based dispatching schemes to overcome the throughput limitation of the RD scheme. First, we introduce a concurrent round-robin dispatching (CRRD) scheme for the Clos-network switch. The CRRD scheme provides high switch throughput without expanding internal bandwidth. CRRD implementation is very simple because only simple round-robin arbiters are adopted. We show via simulation that CRRD achieves 100% throughput under uniform traffic. When the offered load reaches 1.0, the pointers of round-robin arbiters at the first- and second-stage modules are completely desynchronized and contention is avoided. Second, we introduce a concurrent master-slave round-robin dispatching (CMSD) scheme as an improved version of CRRD to make it more scalable. CMSD uses hierarchical round-robin arbitration. We show that CMSD preserves the advantages of CRRD, reduces the scheduling time by 30% or more when arbitration time is significant and has a dramatically reduced number of crosspoints of the interconnection wires between round-robin arbiters in the dispatching scheduler with a ratio of 1//spl radic/N, where N is the switch size. This makes CMSD easier to implement than CRRD when the switch size becomes large.

This paper describes a wavelength-division-multiplexing (WDM) optical switching technique that achieves 5-Tb/s switch throughput. This technique is employed in our previously described OPTically Interconnected Distributed Multi-stage Tb/s-ATM switching Network Architecture, called OPTIMA-2. This is a scalable and non-blocking 3-stage switch employing optical WDM and dynamic bandwidth sharing. An inter-stage, cell-based WDM switch is fabricated in combination with field programmable gate arrays (FPGAs) and 2.5-Gb/s 8-wavelength optical interconnection modules. Experimental results from a 5-Tb/s switching system are introduced.

This paper proposes an innovative pipeline-based maximal-sized matching scheduling approach, called PMM, for input-buffered switches. It dramatically relaxes the timing constraint for arbitration with a maximal matching scheme. In the PMM approach, arbitration operates in a pipelined manner, where K subschedulers are used. Each subscheduler is allowed to take more than one time slot for its matching. Every time slot, one of them provides the matching result. The subscheduler can adopt a pre-existing efficient maximal matching algorithm such as iSLIP and DRRM. PMM maximizes the efficiency of the adopted arbitration scheme by allowing sufficient time for a number of iterations. We show that PMM preserves 100% throughput under uniform traffic and fairness for best-effort traffic of the pre-existing algorithm.

In high-speed and high-capacity packet switches, system reliability is critical to avoid the loss of a huge amount of information and to avoid re-transmission of traffic. We propose a series of concurrent fault-detection mechanisms for a multiple-plane crossbar-based packet switch. Our switch model, called the m + z model, has m active planes and z spare planes. This switch has distributed arbiters on each plane. The spare planes, used for substitution of faulty active ones, are also used in the fault detection mechanism, thus providing sufficient data redundancy for fault detection and location. Our detection scheme is able to detect a single fault in one time slot without increasing transmission overhead. The proposed schemes can be used for switches with different numbers of active planes and the number of spare planes needed for fault detection is small.

We propose a novel architecture, a combined input-crosspoint-output buffered (CIXOB-k, where k is the size of the crosspoint buffer) Switch. CIXOB-k architecture provides 100% throughput under uniform and unbalanced traffic. It also provides timing relaxation and scalability. CIXOB-k is based on a switch with combined input-crosspoint buffering (CIXB-k) and round-robin arbitration. CIXB-k has a better performance than a non-buffered crossbar that uses iSLIP arbitration scheme. CIXOB-k uses a small speedup to provide 100% throughput under unbalanced traffic. We analyze the effect of the crosspoint buffer size and the switch size under uniform and unbalanced traffic for CIXB-k. We also describe solutions for relaxing the crosspoint memory amount and scalability for a CIXOB-k switch with a large number of ports.

A new scheduling algorithm that concatenates data blocks to increase switching throughput is proposed. The algorithm controls the degree of concatenation and reduces the number of switching instances to improve switch utilization. The switch architecture uses virtual output queue switching architecture where the core switch fabric is an optical matrix switch. The optical matrix switch requires the guard-time overhead needed for optical switch control. By reducing the number of switching instances, guard-time overhead can be reduced and switch utilization can be improved. Computer simulations show that efficiency is dramatically increased and that fairness in terms of data throughput among output ports is achieved.

A Clos-network switch architecture is attractive because of its scalability. Previously proposed implementable dispatching schemes from the first stage to the second stage, such as random dispatching, are not able to achieve a high throughput unless the internal bandwidth is expanded. This paper proposes a concurrent round-robin dispatching (CRRD) scheme for a Clos-network switch, to overcome the throughput limitation of the random dispatching scheme. The CRRD scheme provides high switch throughput without expanding internal bandwidth. CRRD implementation is very simple because only simple roundrobin arbiters are adopted. In CRRD, the round-robin arbiters concurrently perform the matching between requesting cells and output links in each first-stage module to dispatch the cells to available second-stage modules. We show that CRRD achieves 100% throughput under uniform traffic. When the offered load reaches 1.0, the pointers of roundrobin arbiters at the first-stage and second-stage modules are effectively desynchronized and contention is avoided. key words: Packet switch, Clos-network switch, dispatching, arbitration, throughput

The hardware architecture of a 5-Tbit/s FB (frame-based) ATM switching system OPTIMA-2 (OPTically Interconnected Distributed Multi-stage Tbit/s-ATM switching Network Architecture-2) is described. OPTIMA-2 has a non-blocking 3-stage switch architecture employing optical WDM links and dynamic bandwidth sharing. The WDM links are composed of a sender port, wavelength arrayed waveguided grating (AWG) router and receiver port. The sender port at the output port of a switch-element allocates a packet to one of eight WDM wavelengths and transmits it. The receiver port receives packets of all eight wavelengths via the wavelength AWG router and merges them at an input port of the next-stage switch-element. Total bandwidth is limited to 10 Gbit/s, but the maximum bandwidth of each wavelength is designed to be 2.5 Gbit/s to prevent the statistical multiplexing gain from falling. A scheduler, which selects variable-length packets from the eight wavelengths, can keep fairness and a small delay. The bandwidth of each wavelength is changed dynamically by the system so that the traffic in each wavelength is equally distributed among the total bandwidth. In OPTIMA-2, variable-length FB-ATM cell, which is familiar with the IP packet, can be switched while keeping the throughput fairness.

This paper proposes a 3-stage ATM switch architecture that uses optical WDM (wavelength division multiplexing) grouped links and dynamic bandwidth sharing. The proposed architecture has two features. The first is the use of WDM technology which makes the number of cables used in the system proportional to system size. The second is the use of dynamic bandwidth sharing among WDM grouped links. This prevents the statistical multiplexing gain offered by WDM from falling even if the switching system becomes large. A performance evaluation confirms the scaleability and cost-effectiveness of the proposed architecture.

An experimental 640-Gbit/s ATM switching system is described. The switching system is scalable and quasi-nonblocking. It uses hardware self-rearrangement in a three-stage network. Hardware implementation results for the switching system is presented. The switching system is fabricated using advanced 0.25 /spl mu/m CMOS devices, high-density multi-chip-module (MCM) technologies, and optical wavelength division multiplexed (WDM) interconnection technologies. A scalable 80-Gbit/s switching module is fabricated combined with a newly developed scalable-distributed-arbitration technique, and a WDM interconnection system that connects all 80-Gbit/s switching modules is developed. Using these components: an experimental 640-Gbit/s switching system is partially constructed. It can be applied to realize future broadband ATM networks.

Proposes active-ATM, a flexible, simple and cost-effective ATM-WAN architecture that can handle multiple user-customized ATM-layer protocols, such as ABR and ABT, by using a simple universal ATM transit network. The architecture enables the construction of flexible networks that can evolve easily. With active-ATM and the ATM multi-protocol emulation network architecture called ALPEN, it is easy to implement new ATM-layer protocols by using user-created programs called active-program capsules that modify only the edge nodes. Because these user-sent program capsules can be used to quickly customize the edge nodes, there is no waiting for standardization and implementation of new services. The ATM-layer protocols are emulated only at the edge nodes, making the transit network independent of customer ATM-layer protocols. The active-ATM edge node is based on the flexible programmable node architecture called PUN (programmable unified node). PUN is a platform for user-programmable ATM-layer services. A prototype system has demonstrated the flexibility of the resulting ATM network. The simple and high-speed universal ATM transit network periodically reports to the edge nodes the performance of the its routes by using periodical route performance check sequences, which are independent of the ATM-layer protocols. The information is used by the edge nodes to manage user cell rates in real time. Flexible, adaptive, and sophisticated protocols that efficiently utilize network resources can be easily supported by the network's edge nodes.