Network slicing in the optical layer enables the creation of multiple virtual network slices on a shared physical infrastructure. When resources are coordinated across multi-domain optical networks, ensuring end-to-end (E2E) performance of the E2E slices is a critical challenge. Due to the autonomy of individual domains and the limited sharing of internal domain information, slice coordinators often face constrained visibility into intra-domain operations. This black-box abstraction of domains makes it difficult to monitor and localize failures from the slice coordinator’s high-level view. In this paper, we address the problem of E2E failure localization under such constraints by focusing on the optimal selection of monitoring paths. We propose a deep reinforcement learning approach to identify a set of E2E monitoring paths that maximizes the capability to localize single-link failures, given a constraint on the number of paths. Our experiments using the GNPy optical network simulator demonstrate that our solution outperforms baseline heuristics and approaches near-optimal performance in localization capability.

Network slicing is a key technological concept for next-generation networking to provide logically dedicated, customized connections for diverse use cases. In many use cases that require access to cloud facilities located far from the network edge, it is crucial to guarantee end-to-end (E2E) performance. However, the composition of E2E network slices demands complex resource coordination among multiple administrative domains that may limit the exposure of network state (e.g., topology, latency) within them for privacy and safety reasons. This paper discusses a resource coordination problem to construct E2E network slices hosted over multiple domains under limited state visibility. The resource coordination task can be formally described as a regret minimization problem with a linear objective function. We present a novel hybrid approach that incorporates partial resource information reported by each domain into the Learning with Linear Rewards (LLR) algorithm. Our experiment results show that the proposed algorithm performs significantly better than other baseline learning algorithms and the LLR algorithm, especially when the traffic patterns of network slices are more dynamic and unstable.

Network slicing enables the deployment of multiple virtual networks over a single infrastructure network, supporting diverse network services. In particular, stringent service requirements motivate End-to-End (E2E) network slicing, where multiple independently administered network domains collectively realize an E2E connectivity. However, because internal resource details of each domain are typically hidden for privacy and administrative reasons, the limited visibility makes accurate E2E performance estimation a significant challenge before its deployment. This paper addresses the problem of truthful information sharing among domain administrators and an E2E slice resource coordinator through a game-theoretic framework. We propose an incentivization strategy based on the Shapley value from cooperative game theory to promote truthful reporting. By quantifying each domain’s contribution to the overall E2E performance, our incentivization strategy adjusts financial compensation to reflect the quality of reported information. Simulations demonstrate that our approach improves E2E performance estimation, reduces performance requirement violations, and lowers slice blocking rates.

Network slicing enables the hosting of heterogeneous services by provisioning virtually isolated networks over a shared network infrastructure. Dynamic network slicing, which allows slices to scale resource usage at runtime, is a promising extension of the original static slicing concept. However, the unpredictable scheduling of resource scaling events can lead to potential blocking, where scale-up requests may be rejected due to resource contention. In this paper, we define a blocking prediction problem, where a slice provider assesses the risk of future blocking, given only the static specifications of slice requests. We propose a Graph Neural Network (GNN)-based solution that generalizes to arbitrary topologies and numbers of requests, predicting blocking events. Our simulation demonstrates that the proposed approach consistently outperforms a threshold-based method, offering a scalable and effective solution for dynamic network slice management.

The development of containers as a tool for scalable computing has led to the increased use of the serverless edge computing paradigm. In particular, containers enable fast deployment of services to edge networks, allowing users to access spatially closer servers and resulting in lower latency. As instantiating containers can cause extra delay called a cold start delay, container caching, which keeps used containers alive in memory for reuse by the next user, has been proposed to mitigate the delay. However, such edge servers are constrained in their memory capacity, and there is a need for an efficient caching strategy. In this paper, we demonstrate that Multi-Agent Reinforcement Learning (MARL) can effectively learn a cache replacement policy and outperform traditional heuristic and centralized Deep Reinforcement Learning (DRL) algorithms. Our results also show that the proposed MARL’s smaller action space has a significant advantage over DRL, which requires full information about the network. In particular, our research demonstrates that MARL is able to scale to larger networks without significant sacrifices to performance.

In response to the security threats posed by Distributed Denial of Service (DDoS) attacks, this paper presents an intrusion detection framework with a high-accuracy multi-class classification model. In addition to the detection of the existence of DDoS attacks, our framework aims to identify a type of attack (e.g., protocol or message type) so that the system can select the most appropriate countermeasure against the DDoS type. We leverage a meta-learner to build an ensemble model of multiple machine learning models such as LSTM, RF, and KNN to enhance detection and classification accuracy. Tested on the CIC-DDoS 2019 dataset, the proposed model archives 96% accuracy in the type identification task, while a simple combination of the existing detection classifiers only achieves 92% accuracy in the same type identification.

Network slicing in Optical Transport Networks (OTNs) is a promising technology to provide hard resource isolation and performance assurance of services with diverse requirements. When network slices are allowed to request their scaling based on their service traffic trends (i.e., dynamic network slicing), it is difficult for a slice provider to maintain resource isolation and high resource utilization simultaneously. We formulate a blocking prediction problem for the slice provider to assess the possibility of future blocking events for a given set of slice requests. We use a Multi-Layer Perceptron classifier to identify a set of slice requests that would cause blocking. It is demonstrated by the simulations with different request patterns that our proposal outperforms other learning-based approaches. The results indicate that a slice provider could assess the risk of experiencing a violation of service assurance when accepting a set of requests and take an appropriate countermeasure such as admission control, using our prediction module.

Serverless edge computing environments use lightweight containers to run different services on a need basis. Container caching at edge nodes is an effective strategy to further reduce the startup latency related to the preparation of container images. However, the capacity limitation of the edge nodes requires an efficient caching strategy that can capture underlying service request patterns. Hence, this paper proposes an EXplainable Reinforcement Learning (XRL)-based container caching strategy to increase the hit rate of cached containers. While a few studies already proposed RL-based caching algorithms, our proposal focuses more on the explainability part of the caching decisions based on a causal graph. The generated explanations from our approach can indicate which caching actions specifically contribute to the increase in the hit rate, which implies the underlying request patterns. Our experiments in a simple network topology demonstrate the validity of the generated explanations.

We address the problem of establishing an end-to-end network slice across multiple domains and propose a Reinforcement Learning-based framework that enables multiple domains to collaborate on end-to-end network slicing admission and allocation. The objective is to maximize the long-term revenue of the network operator. We employ a Graph Neural Network (GNN) to capture the topology features as the encoder. The simulation results show that our framework improves the profit of the network operator by up to 15% compared to a greedy algorithm.

This paper addresses a resource allocation strategy for network slices, where each network slice supports a different federated learning task. A slice is established when a new federated learning model needs to be trained and is released once the training is complete. The goal is to minimize the average network slice holding time while also providing fairness between slice tenants and improving network efficiency. We propose a reinforcement learning-based strategy to periodically reallocate resources according to the current state of each federated learning task. We offer two reinforcement learning models. The first model achieves more stable performance and considers correlations between tasks, while the second model utilizes fewer parameters and is more robust to varying number of tasks. Both approaches have better performance than baseline heuristic methods. We also propose a method to alleviate the effect of various resources scales to make the training stable.

In this work, we consider the network slice composition problem for Service Function Chains (SFCs), which addresses the issue of allocating bandwidth and VNF resources in a way that guarantees the availability of the SFC while minimizing cost. For the purpose of satisfying the availability requirement of the SFC, we adapt a traffic-weighted availability model which ensures that the long-term fraction of traffic supported by the slice topology remains above a desired threshold. We propose a method for composing a single or multi-path slice topology and for properly dimensioning VNF replicas and bandwidth on the slice paths. Through simulations, we show that our proposed algorithm can reduce the total cost of establishment compared to a dedicated protection approach in 5G networks.

The emergence of diverse network applications demands more flexible and responsive resource allocation for networks. Network slicing is a key enabling technology that provides each network service with a tailored set of network resources to satisfy specific service requirements. The focus of this paper is the network slicing of access networks realized by Passive Optical Networks (PONs). This paper proposes a learning-based Dynamic Bandwidth Allocation (DBA) algorithm for PON access networks, considering slice-awareness, demand-responsiveness, and allocation fairness. Our online convex optimization-based algorithm learns the implicit traffic trend over time and determines the most robust window allocation that reduces the average latency. Our simulation results indicate that the proposed algorithm reduces the average latency by prioritizing delay-sensitive and heavily-loaded ONUs while guaranteeing a minimal window allocation to all ONUs.

This paper addresses the problem of admission control for elastic network slices that may dynamically adjust provisioned bandwidth levels over time. When admitting new slice requests, sufficient spare capacity must be reserved to allow existing elastic slices to dynamically increase their bandwidth allocation when needed. We demonstrate a lightweight deep Reinforcement Learning (RL) model to intelligently make ad-mission control decisions for elastic slice requests and inelastic slice requests. This model achieves higher revenue and higher acceptance rates compared to traditional heuristic methods. Due to the lightness of this model, it can be deployed without GPUs. We can also use a relatively small amount of data to train the model and to achieve stable performance. Also, we introduce a Recurrent Neural Network to encode the variable-size environment and train the encoder with the RL model together.

The increasing demand for diverse network services entails more flexible networks that are realized by virtualized network equipment and functions. When such advanced network systems face a massive failure by natural disasters or attacks, the recovery of the entire system may be conducted progressively due to limited repair resources. The prioritization of network equipment in the recovery phase influences the interim computation and communication capability of systems since the systems are operated under partial functionality. Hence, finding the best recovery order is a critical problem, which is further complicated by virtualization due to the interdependence between virtual network functions and infrastructure elements. This paper deals with a progressive recovery problem under limited resources in networks with VNFs, where some interdependencies exist. We prove the NP-hardness of the progressive recovery problem and approach the optimum solution by introducing DeepPR, a progressive recovery technique based on Deep Reinforcement Learning (Deep RL). Our simulation results indicate that DeepPR can achieve near-optimal solutions in certain networks and is more robust to adversarial failures, compared to a baseline heuristic algorithm.

The increasing reliance upon cloud services entails more flexible networks that are realized by virtualized network equipment and functions. When such advanced network systems face a massive failure by natural disasters or attacks, the recovery of the entire system may be conducted in a progressive way due to limited repair resources. The prioritization of network equipment in the recovery phase influences the interim computation and communication capability of systems, since the systems are operated under partial functionality. Hence, finding the best recovery order is a critical problem, which is further complicated by virtualization due to dependency among network nodes and layers. This paper deals with a progressive recovery problem under limited resources in networks with VNFs, where some dependent network layers exist. We prove the NP-hardness of the progressive recovery problem and approach the optimum solution by introducing DeepPR, a progressive recovery technique based on deep reinforcement learning. Our simulation results indicate that DeepPR can obtain 98.4% of the theoretical optimum in certain networks.

Recent advances in the areas of Internet of Things (IoT), big data, and machine learning have contributed to the rise of a growing number of complex applications. These applications will be data-intensive, delay-sensitive, and real-time as smart devices prevail more in our daily life. Ensuring quality of service (QoS) for delay-sensitive applications is a must, and fog computing is seen as one of the primary enablers for satisfying such tight QoS requirements, as it puts compute, storage, and networking resources closer to the user. In this paper, we first introduce FOGPLAN, a framework for QoS-aware dynamic fog service provisioning (QDFSP). QDFSP concerns the dynamic deployment of application services on fog nodes, or the release of application services that have previously been deployed on fog nodes, in order to meet low latency and QoS requirements of applications while minimizing cost. FOGPLAN framework is practical and operates with no assumptions and minimal information about IoT nodes. Next, we present a possible formulation (as an optimization problem) and two efficient greedy algorithms for addressing the QDFSP at one instance of time. Finally, the FOGPLAN framework is evaluated using a simulation based on real-world traffic traces.

The interdependency between different network layers is commonly observed in cyber physical systems and communication networks adopting the dissociation of logic and hardware implementation, such as software defined networking and network function virtualization. This paper formulates an optimization problem to improve the survivability of interdependent networks by restructuring the provisioning relations. A characteristic of the proposed algorithm is that the continuous availability of the entire system is guaranteed during the restructuring of dependencies by the preservation of certain structures in the original networks. Our simulation results demonstrate that the proposed restructuring algorithm can substantially enhance the survivability of interdependent networks and provide insights into the ideal allocation of dependencies.

This paper studies a network design problem to improve the survivability of interdependent networks by restructuring the dependencies. As different types of networked systems become more integrated, the relation between distinct kinds of network devices has become more intertwined. In order to guarantee the robustness of such systems, survivability problems of interdependent networks must be addressed. A characteristic of the proposed algorithm is that the continuous availability of the entire system is guaranteed by the preservation of certain structures in the original networks during the restructuring process. Simulation results demonstrate that the restructuring heuristic can substantially enhance the survivability of interdependent networks.

With the Internet of Things (IoT) becoming a major component of our daily life, understanding how to improve the quality of service for IoT applications through fog computing is becoming an important problem. In this paper, we introduce a general framework for IoT-fog-cloud applications, and propose a delay-minimizing collaboration and offloading policy for fog-capable devices that aims to reduce the service delay for IoT applications. We then develop an analytical model to evaluate our policy and show how the proposed framework helps to reduce IoT service delay.

This paper discusses an election problem of multiple leaders in a network divided into clusters which specifically takes a form of ring structures or cycles. The objective function is defined by considering a positional relation between adjacent leaders aiming at the reduction of communication latency in a distributed network control. We propose several heuristic approaches which can be adopted to solve the optimization problem in a polynomial time. Our simulations verify the effectiveness of the obtained solutions in different types of graph models.

We address the problem of survivable path pair routing in multi- domain optical networks with geographically correlated failures. The objective is to minimize the risk of simultaneous failure of both the primary and backup paths. We develop a probabilistic model to calculate the simultaneous failure probability of both the paths and consider a topology aggregation scheme for domains based on calculating the physical vulnerable overlapping area of two paths within a domain. We develop an inter-domain minimum overlapping area routing algorithm based on the aggregated information from each domain. We compare our algorithm to Suurballe's Algorithm and an optimal approach and we show that our heuristic approach is effective in reducing the total probability of simultaneous failure.

This paper discusses the intractability and heuristics of the Cluster Leader Election Problem (CLEP), which is to assign a controller to one of the switches in each cluster of a network. In order to improve scalability in Software Defined Networking (SDN), a network can be clustered into subnetworks and managed by multiple distributed controllers. CLEP deals with the optimal placement of the distributed controllers in the subnetworks considering metrics between the controllers, such as distance and connectivity. Additionally, it is shown that some metrics within each subnetwork, which are discussed in other literature can be guaranteed by selecting specific clustering methods before the placement.

This paper discusses a composition problem of a resilient tree on a graph modeling a communication network whose links are intermittent. The resiliency of a tree is defined exploiting the binary relation on two kinds of edge weights: availability and commonality. Availability represents a probabilistic stability of a communication link corresponding to an edge, and commonality indicates the influence of an edge based on relative locations of edges on a tree. Our simulation analyzes that the trees satisfying the proposed ordering property on the binary relation provide more stable connections among communication nodes.