The ever-increasing diversity of devices and the complexity of communication content pose significant challenges for network traffic analysis. These limitations can hinder anomaly detection and network control, both essential for guaranteeing stable communication in next-generation networks. This paper proposes an approach leveraging Latent Dirichlet Allocation (LDA) topic modeling for network traffic analysis. LDA, a powerful machine learning technique, excels at identifying hidden thematic structures within data. When applied to network traffic, LDA can differentiate traffic types, such as web browsing, email, or file sharing. To optimize the effectiveness of traffic demand capture, we conducted experiments to determine the appropriate window size for individual connections and the ideal number of topics needed for accurate traffic analysis. Our proposed topic estimation algorithm, utilizing the bag-of-words representation, demonstrates its efficacy in identifying multiple topics within real-world network traffic datasets. This proposed approach has the potential to significantly improve network performance, security, and management.

Deadline-aware job scheduling problems have been attracting attention in the application domains of scientific workflows and commercial data-center wide area networks. Satisfying all of the demands before the corresponding deadlines requires optimally spreading traffic along future times. In this paper, we propose a novel deadline-aware job scheduling method that leverages a deep reinforcement learning policy gradient algorithm. We obtain the following findings through several experiments on evaluating the proposed method. The EDF-trained initial policy with the proposed method outperforms a random initial policy. We evaluate how the training period influence the performance of the proposed method and the relationship between the training period length and the success ratio. We also demonstrate the proposed method outperformed with other job-scheduling methods, i.e., round-robin, first-come-first-served, and earliest-due-first, with the gaps between these methods and the proposed method expanding as the traffic load increased.

It is crucial to identify the cause immediately when a failure occurs at the base station called eNodeB in LTE networks. However, a huge amount of log data generated from the eNodeB prevents the human operator to quickly identify the cause of failure. In order to improve the network operation efficiency, machine learning technique is used to analyze Key Performance Indicator (KPI) data generated from eNodeB and classify the operational status of the eNodeB. However an issue classification with supervised learning requires a large amount of labeled dataset, which takes costly human-labor and time to annotate raw performance metric data. To address this issue, we propose a method that employs Adversarial Autoencoder (AAE), which is a semi-supervised learning method. We evaluate the proposed method using eNodeB log data collected from a service provider LTE network. We confirm that our approach achieves on average 94% accuracy and yields high accuracy even for the class with a small amount of labeled data.

With the increasing network topology complexity and continuous evolution of the new wireless technology, it is increasingly challenging to address the network service outage with traditional methods. In order to improve the network operation efficiency, a suitable machine learning technique is used to learn and classify individual base station into different issue based on multiple performance metrics during a specific time window. However an issue classification with supervised learning requires a large amount of labeled dataset, which takes costly human-labor and time to annotate data. To mitigate the cost and time issues, we propose a method based on few-shot learning that uses Prototypical Networks algorithm to complement the eNodeB states analysis. Using dataset from a live LTE network consists of thousand of eNodeB, our experiment results show that the proposed technique provides high performance while using a low number of labeled data.

As computer networks become vulnerable to attacks from intruders, Intrusion Detection Systems (IDS) is a critical component that monitors activities of computer networks and classifies them as either normal or anomalous. Remarkable advancement of machine learning makes us consider to use supervised machine learning to build IDS. Supervised machine learning requires a large amount of training data, leading to costly human-labor operation; it requires the human operator to examine data, classify them, and annotate them with a label. To address this issue, we propose an IDS that employs semi-supervised learning. Semi-supervised learning uses a small number of labeled data in training dataset to reduce costly human-labor tasks and improves the performance with support of unlabeled data in training dataset. The proposed method employs Adversarial Auto-encoder (AAE), a semi-supervised learning algorithm that incorporates the Generative Adversarial Nets (GAN) into the Auto-encoder (AE). We evaluate the effectiveness of the proposed method using NSL-KDD dataset. We confirm that the proposed method that uses only 0.1 percent of labeled data achieves comparable performance with existing IDSs that use machine learning methods.

With the increasing network topology complexity and continuous evolution of the new wireless technology, it is challenging to address the network service outage with traditional methods. In the long-term evolution (LTE) networks, a large number of base stations called eNodeBs are deployed to cover the entire service areas spanning various kinds of geographical regions. Each eNodeB generates a large number of key performance indicators (KPIs). Hundreds of thousands of eNodeBs are typically deployed to cover a nation-wide service area. Operators need to handle hundreds of millions of KPIs to cover the areas. It is impractical to handle manually such a huge amount of KPI data, and automation of data processing is therefore desired. To improve network operation efficiency, a suitable machine learning technique is used to learn and classify individual eNodeBs into different states based on multiple performance metrics during a specific time window. However, an issue with supervised learning requires a large amount of labeled dataset, which takes costly human-labor and time to annotate data. To mitigate the cost and time issues, we propose a method based on few-shot learning that uses Prototypical Networks algorithm to complement the eNodeB states analysis. Using a dataset from a live LTE network that consists of thousand of eNodeB, our experiment results show that the proposed technique provides high performance while using a low number of labeled data.

In the future mobile communication, communication based on various mobility models is expected. In 5G mobile communication network that can flexibly allocate network resources, it is necessary to predict traffic demands in order to appropriately allocate network resources. Therefore, it is important to predict the behavior of spatio-temporal mobility in order to appropriately allocate network resources. The pervasiveness of mobile devices based services leading to an increasing volume of spatiotemporal datasets and to the opportunity of discovering usable knowledge about mobility behavior. This knowledge is useful to provide stable communication to mobile networks expected to increase traffic flow. In this paper, we propose a method to grasp the behavior of the mobility in spatio-temporal by mining the trajectory data of the mobility obtained from the GPS data to predict the future mobility of the user from frequent patterns. We propose a mining and prediction algorithm that employs the huge amount of trajectory data. We apply sequential pattern mining algorithms including PrefixSpan and BIDE to obtain frequent trajectory patterns from trajectory database. We evaluate the proposed method using actual trajectory dataset, Geolife project, and demonstrate that the proposed method successfully extracts sufficient number of frequent trajectory patterns to predict the future trajectory of mobility.

In distributed SDN, each controller builds the “global network view” by obtaining missing pieces of global network view from other controllers. All SDN controllers need to maintain consistent view of global network in order to make a correct routing decision. An inconsistent global network view may result in incorrect routing decision a switch. In order to achieve correct routing decision while keeping low latency and high throughput of controller, we need to avoid inter-controller communication overhead at the time of flow setup. In order to do so, we need to make an assignment of switches to controllers such that routing decision on each flow can be made by as small number of controllers as possible. In order to achieve such assignment of switches to controllers, we propose a controller placement model based on traffic locality. We identify a cluster of switches such that fair amount of traffic flow remains inside the cluster and assign the switches belonging to the cluster to the same controller. Thereby the inconsistency between controllers has less impact on the accuracy of controllers routing decision for the traffic flow because the controller is like to have all necessary information on the network resource to calculate the route for the traffic flow. In order to make such clusters of switches, we employ Infinite Relational Model (IRM), a nonparametric Bayesian model that discovers relational structure in data sets as a set of clusters. Simulation results show that our proposed method reduces the inconsistent probability that controllers need to update a global network view.

The softwarization of networks is enabled by the SDN (Software-Defined Networking), NV (Network Virtualization), and NFV (Network Function Virtualization) paradigms, and offers many advantages for network operators, service providers and datacenter providers. Given the strong interest in both industry and academia in the softwarization of telecommunication networks and cloud computing infrastructures, a series of special section was established in IEEE Transactions on Network and Service Management, which aims at the timely publication of recent innovative research results on management of softwarized networks.

Softwarization of networks is an important trend, enabled by the NV (Network Virtualization), SDN (Software-Defined Networking), and NFV (Network Function Virtualization) paradigms and offers many advantages for network operators, service providers and datacenter providers. Given the strong interest in both industry and academia in the softwarization of telecommunication networks and cloud computing infrastructures, a series of special issues was established in IEEE Transactions on Network and Service Management, which aims at the timely publication of recent innovative research results on management of softwarized networks.

Modern webpages consist of many rich objects dynamically produced by servers and client terminals at diverse locations, so we face an increase in web response time. To reduce the time, edge computing, in which dynamic objects are generated and delivered from edge nodes, is effective. For ISPs and CDN providers, it is desirable to estimate the effect of reducing the web response time when introducing edge computing. Therefore, in this paper, we derive a simple formula that estimates the lower bound of the reduction of the response time by modeling flows obtaining objects of webpages. We investigate the effect of edge computing in each webpage category, e.g., News and Sports, using data measured by browsing about 1,000 popular webpages from 12 locations in the world on PlanetLab.

With network virtualization, users of network resources, i.e., service providers (SPs), can be separated from the owners of the network resources, i.e., infrastructure providers (InPs). In this case, SPs are not required to own the network infrastructure, and they can focus on providing network services without constructing and managing the network infrastructure. This is expected to encourage small and medium-sized enterprises to start various types of network services. However, to promote the separation between owners and users of network resources, it is important to suitably define the management model of network resources that is necessary to trade resources between InPs and SPs. In this paper, we summarize the requirements for a management model of network resources considering the aspects of both InPs and SPs and propose a management model consisting of allocation blocks (ABs) that satisfies these requirements. We also propose a method of optimally designing the ABs for a given time-series of resource demand by SPs. Through numerical evaluation using the actual log data of a VoD service, we clarify that the proposed method can reduce the total fee paid by SPs compared with a straightforward approach, i.e., uniform ABs.

For Internet service providers to efficiently use network resources, they need to conduct traffic engineering to dynamically control traffic routes to accommodate traffic with limited network resources. The performance of traffic engineering depends on the accuracy of traffic prediction. However, the volume of network traffic has been changing drastically in recent years due to the growth of various types of network services, making traffic prediction increasingly difficult. Our simple ideas to overcome this challenge are to separate traffic into predictable and unpredictable parts and to apply different control policies to predictable and unpredictable traffic. To promote these ideas, we use software-defined networking technology, particularly Open-Flow, that can control macroflows defined by any combination of L2-L4 packet header information such as 5-tuple. In this paper, we therefore propose the macroflow-generating method for separating traffic into predictable macroflows that have little traffic variation and unpredictable macroflows that have large traffic variation within a limited flow table size. We also propose a macroflow-based traffic engineering scheme that uses different routing policies in accordance with traffic predictability. Simulation evaluation results suggest that our proposed scheme can reduce the maximum link load in a network at the most congested time by 34% and the average link load in a network on average by 11% compared with the current traffic engineering schemes.

In current large scale networks, troubleshooting has become more complicated task due to the diversification in the causes of network failures. The increase in the operational costs has become a serious problem. Thus, manualization of the troubleshooting process also becomes important task though it is time-consuming. We propose a method that automatically extracts a workflow for troubleshooting using multiple trouble tickets. Our method extracts an operator's actions from free-format texts and aligns relative sentences between multiple trouble tickets. Finally, we show a novel approach to visualizing a workflow by mining conditional branches using clustering. We validated our method using real trouble ticket data captured from a network operation and showed that it can extract the workflow to identify the cause of failure.

The number of users downloading video content on the Internet has dramatically increased recently, and obtaining huge rich content items, e.g., movie files, is becoming a popular application of the Internet. The transmission bandwidth consumed by delivering rich content is enormous. Therefore, it is urgent for Internet service providers to design an efficient delivery system that minimizes the amount of network resources consumed. Content delivery networks are widely used to reduce the flow hop length and the response time when obtaining content items. To improve the cache efficiency, various methods of replacing caches have been proposed, with the target of improving the cache hit ratio. The effect of reducing the traffic load on network links by delivering content items from caches depends on the distance from the origin server to the users. However, many existing methods do not consider the origin server distance. To maximize the effect of reducing the link load obtained by caches, we propose here a cache-replacement policy based on the hop distance to the origin servers. To simplify the caching algorithm, we divide the storage capacity of cache severs into multiple virtual caches and manage content items separately on the basis of the hop distance to origin servers. We also propose an optimal method for designing the capacities of virtual caches that maximizes the total expected reduction of flow hop length. Through numerical evaluation, we show that the proposed method can decrease the average link load by about 10% compared with the normal least recently used.

There is currently a strong interest in both industry and academia in the softwarization of telecommunication networks and cloud computing infrastructures. This evolution is enabled by three paradigms. First, Software-Defined Networking (SDN) allows network control to be separated from the forwarding plane and allows for a flexible management of the network resources. Second, Network Virtualization (NV) brings virtualization concepts to the network, similar to cloud computing, which was enabled by virtualization of servers. Third, Network Function Virtualization (NFV) focuses on virtualization of software-based network functions. Instead of installing and managing dedicated hardware devices for these functions, they are implemented as software components and deployed on commodity hardware infrastructures, in most cases operated by a network operator or cloud infrastructure provider. Service Function Chaining (SFC) consists of building services using virtual network functions (VNFs). These three paradigms are synergetic and reinforce each other when used together. Several initial SDN, NV and NFV deployments are already operational in providers’ networks.

Web performance, the time from clicking a link on a web page to finishing displaying the web page of the link, is becoming increasingly important. Low web performance of web pages tends to result in the loss of customers. In our research, we measured the time for downloading files on popular web pages by running web browsers on four hosts worldwide using PlanetLab and detected the longest portion in download time. We found the longest portion in download time to be Blocked time, which is the waiting time for the start of downloading in web browsers. In this paper, we propose a method for accelerating web performance by reducing such Blocked time with a cache of rendering results. The proposed method uses an in-network rendering function which renders web pages instead of web browsers. The in-network rendering function also stores the rendering results in its cache and reuses them for other web browsers to reduce the Blocked time. To evaluate the proposed method, we calculated the web performance of web pages whose render results are cached by analyzing the measured download time of actual web pages. We found that the proposed method accelerates web performance of long round trip time (RTT) web pages or long RTT clients if the web pages' dynamic file percentages are within 80%.

Modern websites consist of many rich objects dynamically produced by servers and client terminals at diverse locations. Consequently, we face complications in understanding the communication structure generated when accessing websites. To reduce the response time at browsed websites, many website objects are delivered using content delivery networks (CDNs), in which data objects are delivered from cache servers located close to user terminals. Although the use of CDNs have been assumed to reduce web response time, the actual effect of CDNs on this reduction has not been clarified. To answer this fundamental question, we measured the communication structure of traffic generated when accessing the 1,000 most popular websites from 12 locations worldwide. We found, for example, that it will be desirable to give high priority to entertainment websites at night and to business-related websites during the day.

The transmission bandwidth consumed by delivering rich content is enormous, so it is urgent for Internet service providers to design an efficient delivery system that minimizes the amount of network resources consumed, i.e., minimizing the hop length of delivery flows. Content delivery networks are widely used to reduce the flow hop length and the response time when obtaining content items. To improve the cache efficiency, various methods of replacing caches have been proposed, with the target of improving the cache hit ratio. However, although the effect of delivering content items from caches depends on the distance from the origin server to the users, these methods do not consider the origin server distance. We propose here a cache-replacement policy based on the hop distance to the origin servers. We divide the storage capacity of cache severs into multiple virtual caches and manage content items separately based on the hop distance to origin servers. We also propose an optimal method for designing the capacities of virtual caches that maximizes the total expected reduction of flow hop length. Through numerical evaluation, we show that the proposed method can increase the average expected reduction of flow hop length by about 20% to 80% compared with normal LRU (least recently used).

Traffic engineering (TE) plays an essential role in deciding routes that effectively use network resources. This is particularly important when one considers the increasing time variation of Internet traffic such as streaming and cloud services. Traffic engineering with traffic prediction is one approach to stably accommodating time-varying traffic. This approach calculates routes from predicted traffic to avoid congestion, but predictions may include errors that instead cause congestion. We propose a prediction-based traffic engineering method that is robust to prediction errors by considering the probability distribution of predicted traffic. Our approach is based on a control-theoretic approach called stochastic model predictive control. Routes are calculated using a probability distribution of prediction errors so that the occurrence probability of congestion is lower than an operator-specified level. By considering the multi-step future dynamics of traffic, the routes are changed gradually to avoid route oscillation. We also show a relaxation method for unreliable far-future probabilistic constraints to avoid overly conservative route changes. Through simulations using backbone network traffic traces, we demonstrate that our method can accommodate most traffic variations under a given target link capacity without sudden large routes changes.

Although the use of software-defined networking (SDN) enables routes of packets to be controlled with finer granularity (down to the individual flow level) by using traffic engineering (TE) and thereby enables better balancing of the link loads, the corresponding increase in the number of states that need to be managed at routers and controller is problematic in large-scale networks. Aggregating flows into macro flows and assigning routes by macro flow should be an effective approach to solving this problem. However, when macro flows are constructed as TE targets, variations of traffic rates in each macro flow should be minimized to improve route stability. We propose two methods for generating macro flows: one is based on a greedy algorithm that minimizes the variation in rates, and the other clusters micro flows with similar traffic variation patterns into groups and optimizes the traffic ratio of extracted from each cluster to aggregate into each macro flow. Evaluation using traffic demand matrixes for 48 hours of Internet2 traffic demonstrated that the proposed methods can reduce the number of TE targets to about 1/50 ~ 1/400 without degrading the link-load balancing effect of TE.

Understanding the impacts and patterns of network events such as link flaps or hardware errors is crucial for diagnosing network anomalies. In large production networks, analyzing the log messages that record network events has become a challenging task due to the following two reasons. First, the log messages are composed of unstructured text messages generated by vendor-specific rules. Second, network equipment such as routers, switches, and RADIUS severs generate various log messages induced by network events that span across several geographical locations, network layers, protocols, and services. In this paper, we have tackled these obstacles by building two novel techniques: statistical template extraction (STE) and log tensor factorization (LTF). STE leverages a statistical clustering technique to automatically extract primary templates from unstructured log messages. LTF aims to build a statistical model that captures spatial-temporal patterns of log messages. Such spatial-temporal patterns provide useful insights into understanding the impacts and root cause of hidden network events. This paper first formulates our problem in a mathematical way. We then validate our techniques using massive amount of network log messages collected from a large operating network. We also demonstrate several case studies that validate the usefulness of our technique.

In recent years, the virtual network (VN) for services by IP over WDM has been studied for the flexible and efficient use of network resources. The VNs are constructed by optical paths on the WDM network. Furthermore, by configuring VNs dynamically, resources which have been used for some VN at previous period can be allocated to another VN at the next period, which improves the utilization ratio of network resources. In this paper, we propose the SVN (Static Virtual Network configuration) method and the DVN (Dynamic Virtual network configuration) method for the VN configuration method. We formulate the MILP (Mixed Integer Linear Programming) for the SVN method and the DVN method and evaluate them from viewpoints of resource cost and availability by calculation experiments.

Traffic engineering with traffic prediction is one approach to accommodate time-varying traffic without frequent route changes. In this approach, the routes are calculated so as to avoid congestion based on the predicted traffic. The accuracy of the traffic prediction however has large impacts on this approach. Especially, if the predicted traffic amount is significantly less than the actual traffic, the congestion may occur. In this paper, we propose the traffic prediction methods suitable to the traffic engineering. In our method, we perform preprocessing before the prediction in order to predict the periodical variation accurately. Moreover, we consider the confidence interval for the prediction error and the variation excluded by the preprocessing to avoid the congestion caused by the temporal traffic variation. In this paper, we discuss three preprocessing approaches; the trend component, the lowpass filter, and the envelope. Through simulation, we clarify that the preprocessing by the trend component or the lowpass filter increases the accuracy of the prediction. In addition, considering the confidence interval achieves the lower link utilization within a fixed control period.

Future network should flexibly deal with unexpected network changes caused by the diversification of the network services. In this paper, we propose a managed self-organizing network concept and a dynamic resource allocation mechanism, called as DRAMS. In the managed self-organizing network, multiple virtual networks are accommodated on a single physical network and controlled autonomically based on self-organization. Due to the sharing of limited physical resources, resource contention among virtual networks will be inevitably caused. To cope with this issue, we introduce a dynamic resource allocation to regulate the resource usage of ill-behaved virtual networks. Our research goal is to establish a management mechanism for controllable self-organizing networks. The proposed mechanism allows self-organizing control by virtual networks and just indirectly controls the behavior of virtual network in order to avoid resource contention and improve resource efficiency.

Traffic Engineering (TE) is one efficient approach to handle traffic changes. To perform TE, a server called the Path Computation Element (PCE) collects the traffic information from all nodes within the network. Then, the PCE calculates the routes suitable to the current traffic. However, in a large-scale network, it is difficult for one PCE to collect all traffic information in a short period of time. Thus, it takes time to change the routes according to traffic changes. In this paper, we propose a method that changes the routes suitable to the current traffic soon after the traffic changes. In our method, we hierarchically divide the network into multiple ranges; the ranges of the lowest layer are constructed of a small number of nodes and the ranges of the upper layer are constructed from the multiple ranges of the lower layer. We deploy a PCE for each range. The PCEs in the lowest layer change the routes within a small range in a short interval according to the traffic information within the range to handle the traffic changes that occur in a short period of time. Against the traffic change that cannot be handled in the lower layer, the PCEs in the upper layer change the routes within the large ranges of the upper layer according to the aggregated traffic information collected from the PCEs of the lower layer. We also propose a method to aggregate traffic information and a method to calculate the new routes by using the aggregated traffic information considering the upper bounds of link utilizations. In this method, we aggregate traffic information so that we can calculate the upper bounds of the link utilizations after the route change only from the aggregated traffic information. Then, the PCE obtaining the aggregated traffic information calculates the new routes without causing any new congestion by checking the upper bounds of link utilizations calculated from the aggregated traffic information. In this paper, we evaluate our method by simulation and clarify that our method can mitigate the congestion soon after the traffic changes.

We propose a scalable call admission control (CAC) scheme that avoids link congestion while maximizing the total admissible traffic of an IP backbone network. Our proposed scheme is scalable in the sense that it requires bandwidth reservation management only at ingress nodes instead of at every links in the core network. We have already evaluated the performance of admissible traffic in case where traffic observations are not executed at ingress nodes in the reference [1]. In this paper, in order to increase admissible traffic volumes, we newly formulate an LP problem in case where traffic observations are executed at ingress nodes for estimating offered traffic volumes between ingress and egress node pairs. We evaluated our proposed scheme for NSFNET and COST239 and showed that the total admissible traffic increases about 2.7 and 2.5 times more for NSFNET and COST239, respectively, compared with the case where traffic observations are not executed.

An application-cooperative bandwidth on-demand (BoD) network with dynamic adaptive reconfiguration for Gbps-class HDTV transmission applications is presented. For high-bandwidth data transmission, such as HDTV streaming and Tera-byte-class file transfer, the strong application-network cooperation is one of the key components for facilitating stable and high-performance transmission. The BoD network offers flexible use of high performance IP/optical network resource upon application request and dynamic reconfiguration functionality in response to changes in traffic demands. This paper addresses Gbps-class HDTV transmission over a BoD network based on IP-optical networking, which uses PCE/VMTN architecture for network control, in order to examine application-network cooperation. The design of the BoD network service and its cooperation scheme for HDTV transmission are described. Performance evaluation results and issues of the BoD network and video transmission systems are shown.

A backup topology design algorithm for avoiding congestion in IP fast reroute is presented. In IP fast reroute techniques, detour routes are computed by using backup topologies. Reducing the number of backup topologies is the main problem, but some links will become overloaded if the number of backup topologies is reduced. The proposed backup topology design algorithm splits the traffic on high load links to other links by considering network conditions, such as the traffic matrix or topology. The key idea of the algorithm is a new concept called the special node, which has a higher node degree in a backup topology, to handle the backup topology design problem. The effectiveness of the algorithm in terms of maximum link load reduction is quantitatively shown.

An IP fast reroute control architecture using a centralized control plane where the control plane and data plane are physically separated is presented. In IP fast reroute processes, route computation and failure restoration are independently performed. Our architecture places the route computation function in the central control server to ensure scalability. As for route configuration, the size of the backup routing table is a significant issue in IP fast reroute architecture. We present a backup-table sharing method to reduce its size.

Network virtualization offers users flexibility and controllability in their network. Virtual network users directly customize their own network topologies for a specific purpose by leasing physical resources from physical network providers. However, the autonomic topology managed independently by each user can lead to the instability or performance degradation. We present a framework for policy-based resource management applied to the virtual network environment. Our framework enables sufficient flexibility and controllability for each user; at the same time, the physical network operator can regulate excessive resource usage by the users. We introduce a two-phase resource management model that adds a policy-based resource management function to the existing GMPLS architecture. In addition, we present a formulation of the optimal virtual network design problems reflected by two different requirements of users. We also developed network operation tools to demonstrate our framework.

This paper proposes a link bandwidth design method considering failures in an IP network. When we design the link bandwidth, we need to consider rerouted traffic's bandwidth by failures in order to avoid the traffic congestion. If we knew the point-to-point traffic matrix, we could easily obtain the required link bandwidth after failures. However, the support in routers for measuring traffic matrices is poor and it is generally difficult to measure traffic matrices in a large network. The proposed method does not need traffic matrix measurement; instead, it uses link load measurement before failures which is generally easy to measure. Then, the proposed method calculates the logical least upper bound (in mathematical other words, supremum or sup) of the bandwidth variation at each link by failures by using Linear Programming (LP). Simulation results show that the link bandwidth amount designed by the proposed method is close to the real required amount, which needs traffic matrix measurement. That is, the proposed method provides little wasted bandwidth using link load measurement only. Compared with the simple method which also uses only link load measurement, the proposed method achieves 50% reduction in wasted bandwidth.

Understanding traffic matrices is useful for network design and so forth, but it is difficult to directly measure traffic matrices in IP networks. There is an existing method to estimate traffic matrices by link traffic information which is easy to measure, and routing information. However, this method requires significantly longer computation time as network size grows. This paper proposes large-scale traffic matrix estimation methods using a divide-and-conquer approach. The network is divided into multiple blocks, traffic matrices are estimated per block and estimation results are combined. In terms of division and combination, this paper proposes two detailed methods. By simulation, it is shown that both of the proposed methods can significantly improve computation time, while keeping estimation accuracy of the existing method. In particular, one of the proposed methods can perform traffic matrix estimation within practical computation time in a network of thousands of nodes.

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.

IP fast reroute techniques have been proposed for achieving fast failure recovery in just a few milliseconds. The basic idea of IP fast reroute is to reduce recovery time after failure by precomputing backup routes. A multiple routing configurations (MRC) algorithm has been proposed for obtaining IP fast reroute. MRC prepares backup configurations, which are used for finding a detour route after failure. On the other hand, requiring too many backup configurations consumes more network resources. It is necessary to recover more traffic flows with fewer backup configurations to ensure scalability. We propose a new backup configuration-creation algorithm for maximizing traffic flows which are fast recovered as much as possible under a limited number of backup configurations. The basic idea is to construct a spanning tree excluding failure links with higher link-loads in each backup configuration. We show that our algorithm has more robust on actual large IP networks.

The Internet has grown extremely fast in the last two decades. The number of routes to be supported by the routers has become very large. Moreover, the number of messages exchanged to distribute the routes has increased even faster. To keep up with the increase, network operators regularly have to perform costly upgrades of the routers. It is unclear whether advances in hardware will be able to keep up with the increasing routing load. More importantly, the large number of routes and iBGP messages negatively impacts iBGP convergence time leading to long connectivity losses. In this paper, we propose a scalable way to support the Internet routes in a Service Provider network. We make use of distributed servers that select routes on behalf of the routers. Then, routes are stored in a Distributed Hash Table (DHT). We adapted the concept of DHT for that purpose. Each router maintains its share of Internet routes in addition to a cache of routes currently in use to forward the Internet traffic. We call our proposal SpliTable. We show that our proposal is more scalable in the number of routes supported in each router than current iBGP route distribution solutions. Moreover, the number of control messages exchanged with our proposal is bounded contrary to current sparse iBGP route distribution solutions which may never converge.

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.

This paper describes the implementation and evaluation of layer-1 bandwidth-on-demand (BoD) capabilities in the Japanese academic backbone network, called SINET3. The network has a nationwide GMPLS-based layer-1 platform and provides reservation-based and signaling-based BoD services. The overall architecture for providing BoD services including its capabilities, user interface, path calculation, and interface to drive the layer-1 platform are described. Actual examples of BoD services and evaluations of the path setup/release time in the network are also presented.

This paper describes the network design and configurations of the new Japanese academic infrastructure, called SINET3, which provides a rich variety of network services to more than 700 universities and research institutions. Since the start of full-scale operations in June 2007, the network has expanded its services to include multi-layer transfer services (IP, Ethernet, and layer-1), enriched virtual private network services (L3VPN, L2VPN, VPLS, and L1VPN), enhanced QoS services (packet-based and circuit-based), and brand-new layer-1 bandwidth-on-demand (BoD) services. This paper explains how the network provides these various network services on a single network platform by effectively configuring leading-edge networking components, such as high-performance IP routers, layer- 1 switches, and a BoD server. Evaluations of the network design and configurations confirmed that the networking functions were effectively coordinated. The procedures and techniques related to the configuration validation that covered all phases of the network design and construction are also presented.

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.

This paper describes brand-new layer-1 bandwidth on demand (BoD) services implemented in the new Japanese academic backbone network, called SINET3. SINET3 is an advanced converged network that provides multi-layer transfer, enriched VPN, enhanced QoS, and layer-1 BoD services. The layer-1 BoD services are dynamic layer-1 resource allocation services directly triggered by users and artfully achieved on the multi-service platform by using a layer-1 BoD server. This paper first explains how the network accommodates a wide variety of network services by effectively combining leading-edge technologies. The paper next describes the overall mechanism for the dynamic layer-1 path setup on the multi-service platform and details the functions of the BoD server in many aspects. The designs focus on the tangible achievement of these services over a nationwide network composed of 75 layer-1 switches and 12 IP/MPLS routers.

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.

Although the Internet is playing an increasingly significant role in global communication, it remains vulnerable to malicious traffic such as worms and DoS/DDoS attacks. In the last few years, the emergence of high speed active worms, such as Code Red II, Nimda, SQL Slammer and MS Blaster, has become a serious issue. These worms cause serious damage to communication networks throughout the world by using up network bandwidth. In addition, since conventional firewall systems are located just in front of the server and do not prevent malicious traffic from entering the network, they cannot prevent such network congestion. We have developed a proto-type system of a network firewall using reconfigurable processors. In this paper, we overview the developed system and present its evaluation results.

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.

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.

Many address lookup methods on IP routers have been recently proposed to improve the packet forwarding capability; nevertheless, their performance prediction is very limited because of lack of consideration of actual traffic characteristics in their evaluations. It is necessary to consider actual traffic to predict more realistic performance in routers, especially in the case of layer 3 and 4 switches whose performance is more influenced by flow characteristics. In this paper, we propose new methods for predicting the router's performance based on the statistical analysis of the Internet traffic. We also present an example of its application to the existing table lookup algorithm, and show that simulation results based on our method can provide accurate performance prediction.

This paper proposes IP+ATM switch router architecture for best-effort and guaranteed services for the next generation public data networks. The proposed architecture has advanced features: (1) scalable flow control, (2) scalable packet forwarding performance, (3) QoS control, and (4) a novel table lookup algorithm. Hardware-based high-speed IP packet forwarding engines and a sophisticated ATM switch are combined in the proposed architecture. Distributed packet forwarding engines are interconnected via the core ATM switching fabric, which employs a proprietary flow control mechanism. ATM traffic management and OAM capabilities are fully utilized to provide QoS guarantee for the label switch path (LSP). By combining the high-speed IP packet forwarding and sophisticated ATM LSP control mechanism, we can build the IP+ATM switch router.

This paper proposes a new high-speed network architecture DBTN (dynamic burst transfer time-slot-base network), which is based on circuit-switched network technology. A routing tag is attached to a burst at an ingress edge node and the burst is self-routed in a DBTN network by creating a circuit dynamically. The routing tag, which is called time-slots-relay, is a series of link identifiers to the destination. The time-slots-relay consists of link identifiers from the ingress to the egress nodes and is used to create the circuit. Subsequent data is switched over the circuit being created in an on-the-fly fashion. A link identifier in the time-slots-relay is unique within each node. Each link identifier is loaded into address control memory (ACM) of each circuit-switched transit node, and the circuit to the destination is established. Subsequent user data follows immediately after the time-slots-relay and is sent over the established circuit. Thus short-lived fairly large data transfers like WWW traffic are efficiently carried. A circuit between adjacent nodes is created and released dynamically so bandwidth efficiency is improved compared with conventional circuit-switched networks. Time division multiplexing of the circuit-switched network is utilized so there is no delay jitter or loss. We address the performance of DBTN switches and report on the implementation of the system.

We propose a fast and compact longest match table look-up method for very long network addresses like IP version 6. This method uses two ideas for a routing-table arranged in a tree-structure. The first idea is to make table look-up fast by caching pointers to intermediate nodes in the tree, reducing the number of node traversals. The second idea is to reduce the memory size required for each node in the tree by one-third by eliminating common parts of addresses of adjacent nodes. Evaluating the performance of this method by using actual routing table data of an IP backbone network, we found it was five to ten times faster than a conventional method.

There has been a rapid growth in applications such as World Wide Web browsing, which are characterized by fairly short sessions that transfer substantial amounts of data. Conventional connection-oriented and datagram services are not ideally engineered to handle this kind of traffic. We present a new ATM service, called Dynaflow service, in which virtual circuits are created on a burst-by-burst basis and we evaluate key aspects of its performance. We compare Dynaflow to the fast reservation protocol (FRP) and show that Dynaflow can achieve higher overall throughput due to the elimination of reservation delays, and through the use of shared "burst-stores." We study the queueing performance of the Dynaflow switch and quantify the relationship between the loss ratio and the buffer size.

This paper presents a new communication service for ATM networks that provides one-way, adjustable rate, on-demand communication channels. The proposed dynamic flow service is designed to operate within a multi-service cell switched network that supports both conventional switched virtual circuits and IP packet routing and is designed to complement those services. It is particularly well suited to applications that transmit substantial amounts of data (a few tens of kilobytes or more) in fairly short time periods (up to a few seconds). Much of the current World-Wide Web traffic falls within this domain. Like IP packet networks, the new service permits the transmission of data without prior end-to-end connection establishment. It uses a variant of ATM resource management cells to carry dynamic flow setup information along with the data, including an IP destination address and burst transmission rate. Switches along the path select next an outgoing link dynamically using the burst rate to guide the routing decision. The dynamic flow setup protocol is based on soft-state control and is designed to facilitate the use of buffers that are shared across all ports of a switch, minimizing per port memory costs.

A low-pass filter (LPF) for ATM traffic measurement is presented and an admission control method based on it is proposed. Instantaneous rate is estimated with the LPF and the maximum instantaneous rate (MIR) observed over the monitoring period is used to determine the residual bandwidth in the method. Hence we call it the MIR method. We developed an analytical model to investigate the performance of the MIR method. We confirmed that the MIR method regulates the number of connections properly even for connections with long holding time. In addition, the MIR method is robust against long range dependent traffic, which degrades the robustness of many previously proposed methods. Because the MIR method does not assume any mathematical model and can be implemented with simple hardware, it is well suited for practical ATM switching systems.

A new connection admission control method based on actual virtual path traffic measurements is proposed to achieve high bandwidth efficiency for various types of traffic. The proposed method is based on the measurement of instantaneous virtual path utilization, which is defined as the total cell rate of the active virtual channels normalized by the virtual path capacity. A low-pass filter is used to determine the instantaneous virtual path utilization from crude measurements. A smoothing coefficient formula is derived as a function of the peak rate of the virtual channel. The residual bandwidth is derived from the maximum instantaneous utilization observed during a monitoring period. Simulation shows that the proposed method achieves statistical multiplexing gains of up to 80% of the limit possible with optimum control for similar traffic sources. It can be implemented with very simple hardware. The admission decision is simple: the requested bandwidth is compared with the residual bandwidth. This method is therefore well suited for practical asynchronous transfer mode switching systems.

This paper proposes a rate control technique for high-speed burst transmission with quick response in ALPEN, an ATM multi-protocol emulation network. Under lower load, it is desirable for the network to allow very quick burst access to a lot of bandwidth, like UBR. Under higher load, a mechanism for congestion avoidance or recovery is required, like ABR. The rate control technique proposed provides both properties. In ALPEN, the transit switch sends only information about the residual bandwidth to the source side edge switch. The edge switch responds to user requests quickly because it knows route status. This is very effective because long delay in ATM-WAN reduces response performance. The proposed technique improves ATM-WAN performance because it has a algorithm to collect the information about route status without using any knowledge about delay length. Its performance is confirmed by simulations.

The authors present dynamic call admission control using the distribution of the number of cells arriving during the fixed interval. This distribution is estimated from the measured number of cells arriving at the output buffer during the fixed interval and traffic parameters specified by users. Call acceptance is decided on the basis of online evaluation of the upper bound of cell loss probability, derived from the estimated distribution of the number of calls arriving. QOS (quality of service) standards can be guaranteed using this control when there is no estimation error. The control mechanism is effective when the number of call classes is large. It tolerates loose bandwidth enforcement and loose policing control, and dispenses with modeling of the arrival processes. Numerical examples demonstrate the effectiveness of this control, and implementation is also discussed.>

An exact analysis is presented for the 2*2 switching element with input port buffers that use the cell bypass queueing discipline. An approximate analysis is developed for the Banyan network composed of 2*2 switching elements with buffers that use the cell bypass queuing discipline. An investigation is made of the effect of this discipline on performance through numerical results obtained by analyses as well as simulation results. The authors also examine the nonuniform input traffic models. The analysis shows that the cell bypass queuing discipline considerably improves the performance of the ATM (asynchronous transfer mode) switches based on Banyan networks with respect to both throughput and cell blocking probability.>

An analytical method is developed for a token ring network with single-buffer terminals and an infinite-capacity bridge in which the bridge has priority over the terminals is developed. The priority-mode operation realized by the reservation discipline of IEEE standard 802-5 is considered. An approximate analysis is considered for both terminals and the bridge in two reservation disciplines: the T-reservation discipline in which the reservation field resides in the message trailer, and the H-reservation discipline in which the reservation field resides in the message header. System performance measures (mean waiting times and throughput) derived for these disciplines are compared to those for a mixed discipline which is an alternative method for priority-mode operation in token ring networks. The accuracy of the approximate analytical results is assessed for both terminals and the bridge by comparing them to the simulation results. For almost all tried cases, it is shown to be within 5-20% relative error in the mean message waiting time.>

A token ring network is considered, with priority-mode operation by a reservation discipline which is adopted by an IEEE standard 802.5. In the reservation discipline, a high-priority station can reserve the next transmission right in the reservation field of the message transmitted by the low-priority station when the reservation field passes through the high-priority station. Then the free token at the high-priority level is passed to the high-priority station. After all messages from the high-priority stations are transmitted, the free token is returned to the station in the downstream of the previous low-priority station. To apply this reservation discipline to an interconnected token ring network system, an analytical model is constructed which consists of low-priority stations with single buffer and a single high-priority station with infinite buffer, which is intended to represent a bridge. An exact analysis for the low-priority stations and an approximate analysis for the high-priority station in the reservation discipline is presented. System performance measures (mean waiting times and throughput) derived from the present analysis are compared with a mixed discipline which is an alternative method for priority-mode operation in token ring networks.>