Multi-band elastic optical networks (MB-EONs) transmit information in multiple bands to increase the available capacity. However, they suffer from quality of transmission (QoT) degradation caused by the inter-channel stimulated Raman scattering effect, which requires addressing through tailored resource assignment. Additionally, dynamically arriving and departing optical service requests generate spectrum fragmentation (SF), where spectrum resources become scattered into non-continuous chunks and aggravate service blocking ratio (SBR) even when the total available bandwidth is sufficient. To jointly address these challenges, we propose an SF- and QoT-aware algorithm for routing, band, modulation format, and spectrum assignment (RBMSA), along with proactive spectrum defragmentation (SD), referred to asSFQA-defrag. The algorithm considers SF metrics and QoT levels of available channels across multiple candidate paths to ensure that the QoT requirements are met while minimizing the SF. The SD process proactively reorganizes spectrum allocation to reduce fragmentation by consolidating the spectrum gaps, which leads to lower blocking of future requests. TheSFQA-defragalgorithm is evaluated against benchmark algorithms that independently consider either QoT or SF in three reference backbone topologies. The results demonstrate thatSFQA-defragsignificantly reduces the SBR and SF compared to benchmarks, albeit with a slight increase in the average path length.

As the foundation of global communication networks, optical fibers are vulnerable to various disruptive events, including mechanical damage, such as cuts, and malicious physical layer breaches, such as eavesdropping via fiber bending. Traditional monitoring methods often fail to identify subtle or novel anomalies, stimulating the proliferation of ML techniques for detection of threats before they cause significant harm. In this paper, we evaluate the performance of SSL and USL approaches for detecting various abnormal events, such as fiber bending and vibrations, by analyzing polarization signatures with minimal reliance on labeled data. We experimentally collect thirteen polarization signatures on three different types of fiber cable and process them using OCSVM as an SSL, and DBSCAN as a USL algorithm for anomaly detection. We introduce tailored evaluation metrics designed to guide hyper-parameter tuning and capture generalization over different anomaly types, detection consistency, and robustness to false positives, enabling practical deployment of OCSVM and DBSCAN in optical fiber security. Our findings demonstrate DBSCAN as a strong contender to detect previously unseen threats in scenarios where labeled data are not available, despite some variability in performance between different scenarios, with F1 score values between 0.615 and 0.995. In contrast, OCSVM, trained on normal operating conditions, maintains high F1 scores of 0.98 to 0.998, demonstrating accurate detection of complex anomalies in optical networks.

Reinforcement Learning (RL) has emerged as a powerful tool for automating complex network management tasks, yet its lack of transparency and black-box nature hinder trust and adoption in operational environments. In this work, we focus on explaining the behavior of an $\mathbf{R L}$ agent applied to the problem of network slice admission control. We present a framework that integrates three key components: a Deep Reinforcement Learning (DRL) agent for admission control, an Integer Linear Programming (ILP) model for network slice embedding, and an explanation module for interpreting the DRL agent’s policies, namely Shapley Value Explainable Reinforcement Learning (SVERL). Our analysis aims gives particular attention to cases where the RL agent rejects admitting a network slice request despite sufficient network capacity to provision it, and investigates whether explanations can be used to verify and validate the agent’s behavior prior to deployment approval. Experimental results reveal that the agent’s decisions are primarily influenced by substrate network conditions such as congestion, rather than by the intrinsic characteristics of slice requests. While this conservative policy prevents overload, it also leads to overly cautious rejections. Importantly, the proposed explanation framework provides operators with actionable insights to scrutinize, validate, and refine RL-driven policies before operational deployment.

Integrated access and backhaul (IAB) is one of the promising techniques for 5G networks and beyond (6G), in which the same node/hardware is used to provide both backhaul and cellular services in a multi-hop architecture. Due to the sensitivity of the backhaul links with high rate/reliability demands, proper network planning is needed to ensure the IAB network performs with the desired performance levels. In this paper, we study the effect of infrastructure planning and optimization on the coverage of IAB networks. We concentrate on the cases where the fiber connectivity to the nodes is constrained due to cost. Thereby, we study the performance gains and energy efficiency in the presence of free-space optical (FSO) communication links. Our results indicate hybrid fiber/FSO deployments offer substantial cost savings compared to fully fibered networks, suggesting a beneficial trade-off for strategic link deployment while improving the service coverage probability. As we show, with proper network planning, the service coverage, energy efficiency, and cost efficiency can be improved.

As artificial intelligence and machine learning (AI/ML) models become integral to network operations, their lack of transparency poses a significant barrier to operator trust. Existing explainable artificial intelligence (XAI) techniques often fail to bridge this gap for non-specialists, producing technical outputs that are difficult to translate into actionable insights. This paper presents a framework specifically designed to address this shortcoming. It leverages a moderately sized large language model (LLM) and extends beyond the standard use of SHapley Additive exPlanations (SHAP) feature influence values. The framework employs a structured prompt enriched with mutual feature interaction data to generate human-understandable natural language explanations. To validate our framework, we performed an empirical evaluation on an optical quality of transmission (QoT) estimation use case with human evaluators. We collected independent performance evaluations from specialists, which showed a high inter-evaluator agreement. Compared to a state-of-the-art baseline that uses only SHAP feature influence values in a straightforward prompt, our approach improves the explanation usefulness and scope by$\mathbf{1 2. 2 \%}$and$\mathbf{6. 2 \%}$, while achieving 97.5% correctness.

The operational security of latency-sensitive networked applications is increasingly threatened by evolving malicious attacks that compromise operational integrity and network performance. Human-to-machine (H2M) applications, which rely on seamless bidirectional control signals and haptic feedback transmission, exemplify such latency-sensitive use cases. Existing learning-based malicious attack detection frameworks suffer from their reliance on pre-trained datasets, making machine learning models within them ineffective against previously unseen attack patterns. As attack profiles dynamically evolve, static models become obsolete, necessitating adaptive mechanisms to maintain detection accuracy. In this context, concept drift adaptation will serve as a critical tool for enabling models to continuously adjust to changing traffic distributions and emerging attack patterns. However, real-world H2M applications lack access to accurately labeled malicious traffic data, making real-time adaptation of defense mechanisms infeasible. To address these challenges, we propose a Concept Drift Adaptation-facilitated malicious attack Defense framework (CDAD). Firstly, CDAD employs Adaptive Random Forest as an incremental learning approach, integrating an error-rate-based concept drift detection mechanism to dynamically identify evolving attack patterns and trigger adaptive model updates. Secondly, a haptic behavior classifier is introduced to classify expected human operator interactions and compare them with real-time haptic feedback from remote machines. This enables automated traffic relabeling, allowing CDAD to adapt to previously unseen attacks without relying on pre-labeled datasets. The superior performance of CDAD over existing state-of-the-art methods is demonstrated across various malicious attack scenarios through extensive simulations. Results show that with CDAD, the attack success rate can be limited to 3%, while maintaining an inference time below 1ms, thereby ensuring effective and efficient malicious attack defense in latency-sensitive H2M applications.

The extension of elastic optical network (EON) technologies to multi-band transmission (MB-EON) promises enhanced spectral efficiency, greater throughput, and long-term cost benefits for telecom operators. However, designing such networks presents challenges, particularly in optimizing physical parameters like optical power and quality of transmission (QoT) across different frequency bands. This paper introduces a methodology for optimal span-by-span power allocation using two hyper-accelerated power optimization (HPO) modes: flat launch power (FLP) and flat received power (FRP). This methodology significantly accelerate network power optimization while ensuring service stability in scenarios such as changes in network parameters, QoT degradation due to aging, and network re-optimization or upgrading. Through a comprehensive comparison, we find that FRP notably improves signal flatness and GSNR/OSNR, particularly in the S-band, contributing to a network-wide throughput increase in the order of 12% to 75%. Additionally, we demonstrate that HPO applied to global power optimization is simpler and more cost-effective than when applied to local methods for large-scale networks.

Intelligent computing center (ICC) is a new type of data center constructed with intelligent computing power, such as graphic processing units (GPUs) and artificial intelligence acceleration cards. With billions of parameters, the emergence of large models (e.g., ChatGPT) presents a significant demand of computing power. It may be challenging for a single ICC to provide the required computing power during large model training. Thus, ICC interconnections (ICCI) will become a typical and effective solution to provide intensive computing power. Due to human activities, traditional computing tasks (e.g., transaction processing and online entertainment) exhibit a tidal effect of computing demand, which leads to the tidal variation of remaining computing resources. Moreover, distributed model training (DMT) tasks are likely to cover peaks and valleys of the tidal effect in computing power. In this case, it is easy for DMT tasks to cause an ICC to become a hotspot (i.e., computing load in an ICC exceeds a desired threshold), which significantly degrades the reliability and performance of the ICC. This paper proposes DeepHM, a deep reinforcement learning-based hotspot management strategy through task migration in ICCI networks. To comprehensively consider the bandwidth metrics of the ICCI network, we further propose a dynamic wavelength allocation strategy, i.e., DeepHM-DWA. Simulation results show that the DeepHM and DeepHM-DWA reduce the hotspot compute unit time blocks by 19% and 18% with fewer number of migrated workers while balancing the computing load among multiple ICCs. DeepHM and DeepHM-DWA reduce the average completion time ratio of the DMT tasks by 2% and 5%, respectively.

It is essential to be able to identify hidden anomalies in order to fully automate optical networks. This requires specific features from the application programming interfaces (APIs) used by the control plane and network monitoring solution. One of the solutions, Transport API (TAPI), utilizes advanced techniques in telemetry streaming. The update policy in TAPI enables key performance indicators (KPIs) to be transmitted only when changes are detected. In this paper, we explore how the update policy configuration of TAPI and the use of unsupervised learning (UL) interact in detecting previously unseen anomalies. Results reveal various trade-offs that network operators need to consider, including compute and time overhead, as well as the overall accuracy of UL.

We propose a first Asynchronous Federated Split Learning (AFSL), to add the flexibility of asynchronous computing to the combination of federated and split learning. This amounts to designing a harmonious combination of different paradigms in order to benefit from the advantages of each of them and to reduce the impacts of their shortcomings.This way, AFSL answers to the increasingly rising interest for distributed algorithms with the advent of edge computing in order to support new market segments, such as cloud gaming, immersive eXtended Reality (XR), indoor positioning, and mission critical IoT networks, with stringent requirements on latency and reliability.Computational experiments are conducted on IID and non-IID datasets to investigate the added value of the asynchronous feature. Results indicate that AFSL can accelerate model learning by up to 86% without sacrificing the model’s convergence and accuracy. Indeed, not only average training times are reduced, but clients use fewer resources, a critical characteristic for devices with limited computing capabilities, e.g., in edge devices. Performance degradation can be mitigated by a careful selection of the aggregation principle. Other advantages are with AFSL training in dynamic scenarios as it provides robustness with a short recovery time by leveraging asynchronous client training.

A large language model (LLM) chatbot is integrated within TeraFlowSDN for intent manipulation. The resulting chatbot is capable of understanding the context and is able to carry out three actions: create, find, and explain intents using natural language while being flexible regarding the language used.

Filterless optical networks (FONs) are a cost-effective optical networking technology that replaces reconfigurable optical add-drop multiplexers, used in conventional, wavelength-switched optical networks (WSONs), by passive optical splitters and couplers. FONs follow thedrop-and-wastetransmission scheme, i.e., broadcast signals without filtering, which generates spectrum waste. Programmable filterless optical networks (PFONs) reduce this waste by equipping network nodes with programmable optical white box switches that support arbitrary interconnections of passive elements. Cost-efficient PFON solutions require optimal routing, modulation format and spectrum assignment (RMSA) to connection requests, as well as optimal design of the node architecture. This paper presents an optimization framework for PFONs. We formulate the RMSA problem in PFONs as a single-step integer linear program (ILP) that jointly minimizes the total spectrum and optical component usage. As RMSA is an NP-complete problem, we propose a two-step ILP formulation that addresses the RMSA sub-problems separately and seeks sub-optimal solutions to larger problem instances in acceptable time. Simulation results indicate a beneficial trade-off between component usage and spectrum consumption in proposed PFON solutions. They use up to 64% less spectrum than FONs, up to 84% fewer active switching elements than WSONs, and up to 81% fewer optical amplifiers at network nodes than FONs or WSONs.

We present P5; an event-driven policy framework that allows network operators to realize end-to-end policies on top of P4-based data planes in an intuitive and effective manner. We demonstrate how P5 adheres to a service-level agreement (SLA) by applying P4-based traffic engineering with latency constraints.

Service providers can leverage shared resources to reduce the overall amount of required resources while keeping acceptable Quality of Service (QoS) levels. Kubernetes (K8s) provides a Horizontal Pod Autoscaling (HPA) mechanism that allows to automatically adjust the number of Pods to closely follow the user demand variations over time. To properly leverage shared resources with HPA, service providers need to limit the use of dedicated resources and overprovisioning. However, in the case of traffic spikes, there may not be enough resources to satisfy the demand. The HPA, which relies on resource usage to drive the scaling, is unaware of how many requests could not be served with the required QoS. This might result in an underestimation of the number of required Pods to be added, leading to additional QoS degradation. This demonstration showcases the effectiveness of a new Pod autoscaling mechanism (i.e., Service Aware Pod Autoscaling (SAPA)) that relies on user request measurements from the service load balancer to better estimate the number of required Pods. SAPA allows selecting the amount of Pod resources (dedicated and shared) in a simple way. We demonstrate the benefits of SAPA by comparing it to a K8s cluster based on the traditional HPA in terms of resource usage and service latency.

Exponential growth of bandwidth demand, spurred by emerging network services with diverse characteristics and stringent performance requirements, drives the need for dynamic operation of optical networks, efficient use of spectral resources, and automation. One of the main challenges of dynamic, resource-efficient Elastic Optical Networks (EONs) is spectrum fragmentation. Fragmented, stranded spectrum slots lead to poor resource utilization and increase the blocking probability of in-coming service requests. Conventional approaches for Spectrum Defragmentation (SD) apply various criteria to decide when, and which portion of the spectrum to defragment. However, these polices often address only a subset of tasks related to defragmentation, are not adaptable, and have limited automation potential. To address these issues, we propose DeepDefrag, a novel framework based on reinforcement learning that addresses the main aspects of the SD process: determining when to perform de-fragmentation, which connections to reconfigure, and which part of the spectrum to reallocate them to. DeepDefrag outperforms the well-known Older-First First-Fit (OF-FF) defragmentation heuristic, achieving lower blocking probability under smaller defragmentation overhead.

The ongoing evolution of optical networks towards autonomous systems supporting high-performance services beyond 5G requires advanced functionalities for automated security management. To cope with evolving threat landscape, security diagnostic approaches should be able to detect and identify the nature not only of existing attack techniques, but also those hitherto unknown or insufficiently represented. Machine Learning (ML)-based algorithms perform well when identifying known attack types, but cannot guarantee precise identification of unknown attacks. This makes Root Cause Analysis (RCA) crucial for enabling timely attack response when human intervention is unavoidable. We address these challenges by establishing an ML-based framework for security assessment and analyzing RCA alternatives for physical-layer attacks. We first scrutinize different Network Management System (NMS) architectures and the corresponding security assessment capabilities. We then investigate the applicability of supervised and unsupervised learning (SL and UL) approaches for RCA and propose a novel UL-based RCA algorithm called Distance-Based Root Cause Analysis (DB-RCA). The framework’s applicability and performance for autonomous optical network security management is validated on an experimental physical-layer security dataset, assessing the benefits and drawbacks of the SL- and UL-based RCA. Besides confirming that SL-based approaches can provide precise RCA output for known attack types upon training, we show that the proposed UL-based RCA approach offers meaningful insight into the anomalies caused by novel attack types, thus supporting the human security officers in advancing the physical-layer security diagnostics.

The operation and management of software-based communication systems and services is a big challenge for infrastructure and service providers. The challenge is mainly associated with the more significant number of configurable elements and the higher dynamicity in the software-based systems than the classical ones. On the other hand, the modularity and programmability in software-based networks enabled by technologies like Software-Defined Networking (SDN) and Network Function Virtualization (NFV) provide new opportunities for operators to realize advanced network and service management strategies beyond the classical techniques. In our work, we elaborate on these new opportunities and propose a novel strategy for the management of survivable cloud services. In particular, we leverage the flexibility of SDN and NFV to combine proactive protection and reactive restoration mechanisms, and we put forward a novel strategy for enhancing the survivability of cloud services. Through comprehensive evaluations, we demonstrate that the proposed strategy offers significant benefits in terms of availability and restorability of services while reducing, at the same time, the overhead caused by the relocation of cloud services in case of failures.

Abstract High throughput, resilience, and low latency requirements drive the development of 5G‐enabled content delivery networks (CDNs) which combine core data centers (cDCs) with edge data centers (eDCs) that cache the most popular content closer to the end users for traffic load and latency reduction. Deployed over the existing optical network infrastructure, CDNs are vulnerable to link cut attacks aimed at disrupting the overlay services. Planning a CDN to balance the stringent service requirements and increase resilience to attacks in a cost‐efficient way entails solving the content placement problem (CPP) across the cDCs and eDCs. This article proposes a framework for finding Pareto‐optimal solutions with minimal user‐to‐content distance and maximal robustness to targeted link cuts, under a defined budget. We formulate two optimization problems as integer linear programming (ILP) models. The first, denoted as K‐best CPP with minimal distance (K‐CPP‐minD), identifies the eDC/cDC placement solutions with minimal user‐to‐content distance. The second performs critical link set detection to evaluate the resilience of the K‐CPP‐minD solutions to targeted fiber cuts. Extensive simulations verify that the eDC/cDC selection obtained by our models improves network resilience to link cut attacks without adversely affecting the user‐to‐content distances or the core network traffic mitigation benefits.

With advances in the information and communication technology (ICT), connected vehicles are one of the key enablers to unleash intelligent transportation systems (ITS). On the other hand, the envisioned massive number of connected vehicles raises the need for powerful communication and computation capabilities. As an emerging technique, fog computing is expected to be integrated with existing communication infrastructures, giving rise to a concept of fog-enhanced radio access networks (FeRANs). Such architecture brings computation capabilities closer to vehicular users, thereby reducing communication latency to access services, while making users capable of sharing local environment information for advanced vehicular services. In the FeRANs service migration, where the service is migrated from a source fog node to a target fog node following the vehicle's moving trace, it is necessary for users to access service as close as possible in order to maintain the service continuity and satisfy stringent latency requirements of real-time services. Fog servers, however, need to have sufficient computational resources available to support such migration. Indeed, a fog node typically has limited resources and hence can easily become overloaded when a large number of user requests arrive, e.g., during peak traffic, resulting in degraded performance. This paper addresses resource management in FeRANs with a focus on management strategies at each individual fog node to improve quality of service (QoS), particularly for real-time vehicular services. To this end, the paper proposes two resource management schemes, namely fog resource reservation and fog resource reallocation. In both schemes, real-time vehicular services are prioritized over other services so that their respective vehicular users can access the services with only one hop. Simulation results show that the proposed schemes can effectively improve one-hop access probability for real-time vehicular services implying low delay performance, even when the fog resource is under heavy load.

This paper targets the lifetime-aware management of a set of optical line amplifiers (OLAs) in an optical network exploiting sleep mode (SM) in order to save energy. We first present a simple model to predict the OLA lifetime. We then provide different mixed-integer linear programming (MILP) formulations, which jointly consider energy saving and lifetime. The proposed MILP formulations are then solved on different realistic scenarios, by taking into account the spatial and temporal variations of traffic demands. Results show that our lifetime-aware approach outperforms classical energy saving ILP formulations, which instead tend to notably decrease the OLA lifetime. More important, the proposed approaches can achieve a good lifetime performance without consuming significantly more energy than purely energy-aware strategies.

Service relocation represents a promising strategy to provide flexible and resource efficient resiliency from link failures in the optical cloud environment. However, when a failure affects a node hosting a datacenter (DC), service relocation from the affected DC is not possible. One alternative to protect against DC failures relies on using design strategies that duplicate the IT (i.e., storage and processing) resources in a backup DC at the expense of increasing resource overbuild (i.e., cost) of the network. This work proposes a dimensioning strategy based on the shared-path shared-computing (SPSC) concept able to protect against any single link, server, or DC failure scenario with minimal resource overbuild for the network and IT infrastructures. SPSC is based on the intuition that only storage units need complete replication in backup DC, while processing units can be instantiated only after the occurrence of a failure, leaving the design strategy some leeway to minimize their number. As result, the proposed SPSC design shows a considerable reduction in the amount of backup resources when compared to the dedicated protection strategies.