The increasing availability of AI-based tools such as ChatGPT has transformed how students conduct online searches. This study investigates how students search for information with and without generative AI assistance, how distinct behavioural patterns emerge, and how these patterns relate to performance outcomes. We analyse 726 search stories, i.e., their navigation actions, from 305 students completing up to three search tasks. We first examine performance differences across search modalities, comparing AI-assisted and traditional web search approaches. We then analyse students’ search behaviours and examining their association with performance outcomes. Preliminary results show that AI-based searching tends to cluster around specific behavioural combinations and reduces extreme errors, but does not systematically improve correctness compared to traditional web searches. Furthermore, while performance outcomes vary across different search behaviours, no consistent patterns emerge linking specific behaviours to stable performance across tasks.

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.

Video-based communication is central to today’s digital society. While YouTube and Netflix once dominated video traffic, traditional broadcasters now run their own streaming services, and live-streaming platforms are expanding. Users expect high-resolution video with minimal buffering and latency, especially for live content. Meeting these demands requires infrastructure that balances performance, cost, energy, and resources, supported by comprehensive traffic monitoring and analysis. Artificial Intelligence plays a key role, particularly in encrypted traffic classification and quality prediction, with treebased Machine Learning models like Random Forests (RF) widely used. However, research often emphasizes small accuracy gains while overlooking the raising energy and resource costs of such models, an increasing conflict with sustainability goals. To study this trade-off, we implement RF models with varying feature counts and complexities to predict video re-buffering and buffer health, two key quality indicators. Using a dataset of more than 11,000 YouTube sessions, we analyze how prediction performance scales with model complexity and energy consumption across the full pipeline, revealing several unexpected results.

Model calibration and feature selection are two critical aspects in developing reliable and accurate machine learning models. Calibration ensures that the model's confidence scores accurately reflect the true likelihood of correctness, which is essential in security-critical applications, like Network Intrusion Detection Systems (NIDS). Feature selection (FS), meanwhile, enhances model efficiency, interpretability, and generalization by identifying the most relevant inputs. However, it may also degrade model's calibration if not explicitly considered. Recently, Explainable Artificial Intelligence (XAI) methods, particularly Shapley Additive Explanations (SHAP), have proven effective in guiding the FS process. Yet, existing SHAP-based FS techniques typically focus on improving accuracy, often giving little attention to the impact of FS on model calibration, an aspect that is critical for reliable decision-making in high-stakes applications such as NIDS. In this work, we propose a novel approach that incorporates a calibration-aware loss function for XGBoost with SHAP-based Recursive Feature Elimination for FS to jointly improve both predictive accuracy and model calibration. Experiments on two benchmark NIDS datasets show that our approach can reduce Brier score and Expected Calibration Error by up to 3.5% and 84.8%, respectively, over uncalibrated baselines, and by up to 2.6% and 58.3% over standard calibration methods, while enhancing or maintaining predictive accuracy and number of features.

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.

Network slicing, a key technology for 6G communications, allows diverse services to coexist on a shared physical infrastructure by allocating different resources to virtual networks (VNs, or equivalently, “network slices”) mapped over the shared infrastructure. However, it presents challenges in terms of failure survivability, as the failure of one physical element can lead to the failure of multiple VNs mapped to it, making survivability of ultra-reliable services against multiple failures a crucial research topic. In this study, we investigate the Survivable Virtual Network Mapping (SVNM) problem, focusing on double-link failures. SVNM against double-link failures can be guaranteed by enforcing appropriate SVNM constraints (e.g., any double-link failure cannot disconnect any virtual node from other virtual nodes in the same VN), but this approach requires excessive redundant capacity deployment. To address this issue, we propose a novel technique called SVNM with Inter-VN Capacity Sharing (SINC), which allows capacity sharing across different VNs to improve survivability against double-link failures with efficient spare capacity utilization. Since SINC may fail to reconnect some VNs due to insufficient spare capacity, we propose combining it with a spare slice (a VN fully dedicated to enhancing survivability) to create an advanced version, SINC+, which improves survivability by reconnecting VNs with additional spare capacity. We then formulate both SINC and SINC+ through Integer Linear Programming (ILP) models, which can provide optimal solutions. Moreover, to address the computational limitations of the ILP formulation, we developed scalable heuristic algorithms applicable to both SINC and SINC+ with a small optimality gap. Our numerical results show that VN availability using SINC improves by up to 9.48% over SVNM, with the same total link resource consumption (TLRC). Furthermore, SINC+ ensures VN survivability against all potential double-link failures, and the additional TLRC can be reduced to less than 1% in presence of a high number of VNs or large nodal connectivity, underscoring the sustainability of our proposed solutions.

We investigate classifying hardware failures in microwave networks via Machine Learning (ML). Although MLbased approaches excel in this task, they usually provide only hard failure predictions without guarantees on their reliability, i.e., on the probability of correct classification. Generally, accumulating data for longer time horizons increases the model’s predictive accuracy. Therefore, in real-world applications, a trade-off arises between two contrasting objectives: i) ensuring high reliability for each classified observation, and ii) collecting the minimal amount of data to provide a reliable prediction. To address this problem, we formulate hardware failure-cause identification as an As-Soon-As-Possible (ASAP) selective classification problem where data streams are sequentially provided to an ML classifier, which outputs a prediction as soon as the probability of correct classification exceeds a user-specified threshold. To this end, we leverage Inductive and Cross Venn-Abers Predictors to transform heuristic probability estimates from any ML model into rigorous predictive probabilities. Numerical results on a real-world dataset show that our ASAP framework reduces the time-to-predict by 8x compared to the state-of-the-art, while ensuring a selective classification accuracy greater than 95%. The dataset utilized in this study is publicly available, aiming to facilitate future investigations in failure management for microwave networks.

Quantum Key Distribution (QKD) is currently being explored as a solution to the threats posed to current cryptographic protocols by the evolution of quantum computers and algorithms. However, single-photon quantum signals used for QKD permit to achieve key rates strongly limited by link performance (e.g., loss and noise) and propagation distance, especially in multi-node QKD networks, making it necessary to design a scheme to efficiently and timely distribute keys to the various nodes. In this work, we introduce the new problem of joint Routing, Channel, Key-rate and Time-slot Assignment (RCKTA), which is addressed with four different network settings, i.e., allowing or not the use of optical bypass (OB) and trusted relay (TR). We first prove the NP-hardness of the RCKTA problem for all network settings and formulate it using a Mixed Integer Linear Programming (MILP) model that combines both quantum channels and quantum key pool (QKP) to provide an optimized solution in terms of number of accepted key rate requests and key storing rate. To deal with problem complexity, we also propose a heuristic algorithm based on an auxiliary graph, and show that it is able to obtain near-optimal solutions in polynomial time. Results show that allowing OB and TR achieves an acceptance ratio of 39% and 14% higher than that of OB and TR, respectively. Remarkably, these acceptance ratios are obtained with up to 46% less QKD modules (transceivers) compared to TR and only few (less than 1 per path) additional QKD modules than OB.

Quantum Key Distribution (QKD) is a recent technology for secure distribution of symmetric keys, which is currently being deployed to increase communications security against quantum attacks. However, the key rate achievable over a weak quantum signal is limited by the link performance (e.g., loss and noise) and propagation distance, especially in multi-node QKD networks, making it necessary to design a scheme to efficiently and timely distribute keys to the various nodes. In this work, we formulate, using a Mixed Integer Linear Programming (MILP) model, a novel Routing, Channel, and Key-rate Assignment (RCKA) problem for QKD with Quantum Key Pool (QKP), which exploits the opportunity of using trusted relays and optical bypass. Our formulation accounts for the possibility to build a quantum key distribution path that combines both quantum channels and trusted relays to increase the acceptance ratio of key rate requests. Leveraging different versions of the proposed MILP model, we evaluate several strategies exploiting different combinations of trusted relays and optical bypass for the RCKA problem. Results show how different trade-offs between security and resource-efficiency (expressed in terms of acceptance ratio of key rate requests vs. key storing rate in QKP) can be achieved when adopting trusted-relay and/or optical-bypass technologies. Trusted relays can provide a higher acceptance ratio when the number of QKD modules (transmitters or receivers) is sufficiently large, while optical bypass, which does not require the implementation of expensive trusted relays, is preferable when the number of QKD modules is a limiting factor.

Artificial Intelligence (AI) has demonstrated superhuman capabilities in solving a significant number of tasks, leading to widespread industrial adoption. For in-field network-management application, AI-based solutions, however, have often risen skepticism among practitioners as their internal reasoning is not exposed and their decisions cannot be easily explained, preventing humans from trusting and even understanding them. To address this shortcoming, a new area in AI, called Explainable AI (XAI), is attracting the attention of both academic and industrial researchers. XAI is concerned with explaining and interpreting the internal reasoning and the outcome of AI-based models to achieve more trustable and practical deployment. In this work, we investigate the application of XAI for network management, focusing on the problem of automated failure-cause identification in microwave networks. We first introduce the concept of XAI, highlighting its advantages in the context of network management, and we discuss in detail the concept behind Shapley Additive Explanations (SHAP), the XAI framework considered in our analysis. Then, we propose a framework for a XAI-assisted ML-based automated failure-cause identification in microwave networks, spanning model’s development and deployment phases. For the development phase, we show how to exploit SHAP for feature selection and how to leverage SHAP to inspect misclassified instances during model’s development process, and how to describe model’s global behavior based on SHAP’s global explanations. For the deployment phase, we propose a framework based on predictions uncertainty to detect possibly wrong predictions that will be inspected through XAI.

Filterless Optical Networks (FONs) (i.e., optical networks where switching nodes are solely based on passive splitters and combiners) enjoy features that are highly appreciated by network operators, such as their low cost and their energy efficiency, posing them as an alternative solution to filtered Wavelength-Switched Optical Networks (WSON) based on active switching nodes. Due to FONs’ specific design criteria (the network topology must be divided into link-disjoint filterless fiber trees to avoid laser loops), traditional network problems, such as survivable virtual network mapping, shall be revisited and tackled adopting novel solutions with respect to state-of-the-art filtered WSONs. In this paper, we investigate the problem of survivable virtual network mapping (SVNM) in FONs with the aim of evaluating the cost of survivability when adopting FON technology. We first model the problem as an Integer Linear Program to establish fiber trees and provide survivable mapping of virtual networks, while minimizing cost of additional network equipment and spectrum with respect to WSON. We then propose multiple heuristic and meta-heuristic approaches to tackle large problem instances. In our numerical evaluations, we consider three scenarios: FON, WSON, and FON with pre-established fiber trees. Results show that in FON, where SVNM is jointly optimized with fiber tree establishment, the investment in additional network equipment can be largely minimized, and even avoided in some cases. In contrast, in FON with pre-established trees, amount of additional network equipment needed to guarantee survivability is significant (up to 60% with respect to WSON).

In presence of multiple failures affecting their network infrastructure, operators are faced with the Progressive Network Recovery (PNR) problem, i.e., deciding the best sequence of repairs during recovery. With incoming deployments of 5G networks, PNR must evolve to incorporate new recovery opportunities offered by network slicing. In this study, we introduce the new problem of Progressive Slice Recovery (PSR), which is addressed with eight different strategies, i.e., allowing or not to change slice embedding during the recovery, and/or by enforcing different versions of slice connectivity (i.e., network vs. content connectivity). We propose a comprehensive PSR scheme, which can be applied to all recovery strategies and achieves fast recovery of slices. We first prove the PSR’s NP-hardness and design an integer linear programming (ILP) model, which can obtain the best recovery sequence and is extensible for all the recovery strategies. Then, to address scalability issues of the ILP model, we devise an efficient two-phases progressive slice recovery (2-phase PSR) meta-heuristic algorithm, small optimality gap, consisting of two main steps: i) determination of recovery sequence, achieved through a linear-programming relaxation that works in polynomial time; and ii) slice-embedding recovery, for which we design an auxiliary-graph-based column generation to re-embed failed slice nodes/links to working substrate elements within a given number of actions. Numerical results compare the different strategies and validate that amount of recovered slices can be improved up to 50% if operators decide to reconfigure only few slice nodes and guarantee content connectivity.

Network Slicing is one of the key enabling technologies in 5G networks, as it allows the same network infrastructure to host numerous services, characterized by different Quality of Service (QoS) requirements. Network slicing provides greater flexibility when assigning resources to virtual networks (VNs, or, equivalently, “network slices”), allowing to meet very diverse service requirements. However, network slicing also brings numerous challenges in terms of management of network resources. Among these, service reliability is one of the most important, especially in light of the rising importance of ultra-reliable services in 5G. In this study, we investigate the Survivable Virtual Network Mapping (SVNM) problem focusing on double-link failures. SVNM against double-link failures can be guaranteed enforcing appropriate SVNM constraints, but this approach requires excessive redundant capacity. Capacity sharing represents a more capacity-efficient solution to ensure survivability against double-link failures. Hence, we propose a new SVNM strategy that allows capacity sharing across different virtual networks in case of double-link failure. To evaluate benefits of the proposed technique we categorize six different SVNM scenarios (with and without capacity sharing, jointly applied with SVNM or not) and formalize them through Integer Linear Programming (ILP) models. Results show that the proposed technique for SVNM with capacity sharing enables availability gains (up to about 29%) over traditional SVNM against single-link failures and significant capacity savings (up to about 50%) over SVNM against double-link failures. The advantages are more significant for increasing number of virtual networks.

Enabling Dedicated Path Protection (DPP) in Filter-less Optical Networks (FONs) poses specific design challenges, as FONs require dividing the network topology in non-overlapping fiber trees, and lightpaths cannot cross from one tree to another unless additional devices are installed. In this study, we consider the possibility to deploy three type of devices, namely I nter-Tree Transceivers (ITTs), Wavelength Blockers (WBs) and Colored Passive Filters (CPFs) to achieve DPP in FON, and we compare the three resulting DPP strategies, called P-ITT, P- WB and P- WBC. More specifically, we formulate three Integer Linear Programming (ILP) models for DPP in FON with the objective to minimize additional device cost and minimize total wavelength consumption. Numerical results over two realistic topologies show that P-WBC achieves cost savings up to 33% in comparison to P-WB and up to 97% in comparison to P-ITT. However, even if it is the costliest approach, P-ITT ensures up to 7 % savings in wavelength consumption and up to 23 % savings in resource overbuild compared to P- WB and P- WBC, making it a possible candidate in spectrum-scarce deployments.

Network Function Virtualization (NFV) provides a major shift in the provisioning of telecommunication services by decoupling network functions from dedicated hardware devices. Such decoupling enables operational expenditure (OpEx) and capital expenditure (CapEx) reduction and allows to increase service agility. NFV relies on Virtualized Network Functions (VNFs) and, by placing VNFs on NFV-capable network nodes, and by chaining them in a specific order while guaranteeing a given end to end latency, Service Chains (SCs) are formed to provide a specific service. To achieve great flexibility in resource assignment in the network and decrease further OpEx, it is important to consider provisioning of SCs in a dynamic scenario in which traffic evolves in the network. In this study we observe that, when deploying a SC in a situation where SC requests arrive dynamically in the network, it is important to consider protection techniques to withstand failures of the network components supporting the SC. Different protection approaches can be followed to protect the SC against failures. We consider three different protection strategies, namely, Virtual-Node protection, Virtual-Link protection and End-to-End protection, which provide protection against single virtual node (hosting a VNF), single virtual link (connecting two consequent VNF of SC together) and single virtual node/virtual link failure for dynamic VNF placement. For each of them, we provide a heuristic approach for dynamic provisioning of the SC with protection. In our simulative numerical results over realistic network and SC settings, we compare the three strategies and show that End-to-End protection and Virtual-Node protection have both high blocking, however, End-to-End is able to satisfy the latency requirement of more SCs with respect to virtual node protection. Of the three protection strategies, Virtual-Link protection requires less network and computational resources and achieves lower SC latency violation.

With the risk of natural disaster occurrence rising globally, the interest in innovative disaster resilience techniques is greatly increasing. In particular, Data Center (DC) operators are investigating techniques to avoid data-loss and service downtime in case of disaster occurrence. In cloud DC networks, DCs host Virtual Machines (VM) that support cloud services. A VM can be migrated, i.e., transferred, across DCs without service disruption, using a technique known as “online VM migration”. In this article, we investigate how to schedule online VMs migrations in an alerted disaster scenario (i.e., for those disasters, such as tsunami and hurricanes, that grant an alert time to DC operators) where VMs are migrated from a risky DC, i.e., a DC at risk to be affected by a disaster, to a DC in safe locations, within a deadline set by the alert time of the incoming disaster. We propose a multi-objective Integer Linear Programming (ILP) model and heuristic algorithms for efficient online VMs migration to maximize number of VMs migrated, minimize service downtime and minimize network resource occupation. The proposed approaches perform scheduling, destination DC selection and assign route and bandwidth to VM migrations. Compared to baseline approaches, our proposed algorithms eliminate service downtime in exchange of an acceptable additional network resource occupation. Results also give insights on how to calculate the minimum amount of time required to evacuate all VMs with no service downtime. Moreover, since the proposed approaches exhibit different execution times, we design an ‘alert-aware VM evacuation’ tool to intelligently select the most suitable approach based on the number and size of VMs, alert time and available network capacity.

Automated failure-cause identification in communication networks allows operators to reduce service unavailability. Once the most likely failure root-cause is identified, appropriate countermeasures can be effectively put in place (e.g., by choosing an in-field intervention vs. a remote equipment reconfiguration). In this article, we describe a successful application of Machine Learning (ML) for automatic failure identification in microwave networks based on the real-field data. On microwave links, different heterogeneous causes (e.g., adverse atmospheric conditions, or obstacles) lead to service unavailability and produce not easily-distinguishable degradation effects on the transmission parameters. Hence, failure identification is traditionally accomplished by domain experts via direct inspection of transmission-parameter logs. As a first contribution, we identify six categories of failure causes in microwave networks and show that supervised ML enables very accurate failure identification, hence significantly simplifying failure troubleshooting. Comparing various ML algorithms, we find that up to 93% classification accuracy is obtained using real-field labeled datasets with 2513 points. One main hindrance to the application of supervised learning is that, in real network deployments, limited amount of labeled data is available for training, as manual labeling is performed by domain experts based on their knowledge and experience. On the other hand, collecting unlabeled data is relatively simple as network management systems retrieve large amounts of unlabeled information automatically. As a second contribution, we investigate an automated labeling procedure, based on autoencoders-like Artificial Neural Networks, to combine the knowledge of the few manually-labeled data with large unlabeled data. Results show that our data augmentation based on autoencoders can slightly improve failure-cause identification only when Artificial Neural Networks or Support Vector Machines are used, while accuracy slightly decreases when adopting Random Forest.

Network Function Virtualization (NFV) refers to the process of running network functions in virtualized IT infrastructures as softwarized Virtual Network Functions (VNFs). Several telecom service providers are currently benefiting from this concept, as it enables a faster introduction of new network services, thereby meeting changing requirements. Following a trend initially adopted by cloud service providers, telecom service providers are also adopting de-aggregation of the VNFs into microservices (μservices). However, a μservice-based architecture that can manage a large set of diverse and sensitive network functions requires new Artificial Intelligence (AI)-based methodologies to cope with the complexity of the μservice-based NFV paradigm. This paper focuses on the use of explainable AI (XAI) for gradually migrating towards a μservices-based architecture in NFV. The paper first establishes the need for XAI to transform the NFV architecture to a μservice-based architecture and then describes some of our research objectives. Afterwards, our preliminary approach and long-term visions are provided.

In Filterless Optical Networks (FONs), passive splitters and combiners replace the more complex and more expensive Reconfigurable Optical Add-Drop Multiplexers (ROADMs) based on Wavelength Selective Switch (WSS) in network nodes. The utilization of passive switching elements in network nodes requires to subdivide the network topology into loop-free edge-disjoint fiber trees, consequently constraining physical paths between nodes. In this work, we investigate Virtual Network Mapping (VNM) and Virtual Network Embedding (VNE) with virtual link protection in the context of FONs. To ensure virtual link protection, we propose a strategy based on the placement of additional network equipment to allow mapping virtual links over two fiber trees. We model the problem as an Integer Linear Programming (ILP) formulation with the objective of minimizing network cost in terms of additional network equipment and overall wavelength consumption. Main results show that, when solving VNE, placement of virtual nodes and link mapping can be optimized to avert excessive wavelength consumption, while VNM drains much more network's capacity, since virtual nodes' locations are pre-determined. In addition, results show that, when guaranteeing protection of virtual links in FONs, the gain of VNE is further maximized with respect to VNM. Moreover, results show that solving VNM with virtual link protection requires additional network equipment, severely penalizing network cost.

In view of incoming 5G mobile communication, network operators must upgrade their network capacity while capping capital and operational expenditures. Filterless Optical Networks (FONs) are emerging as a cost-effective technology as they eliminate costly active switching elements, the Reconfigurable Optical Add-Drop multiplexers (ROADMs) based on Wavelength Selective Switch (WSS), by replacing them with passive devices as optical power splitters/combiners. However, eliminating active switching and filtering components enforces signal broadcast on all the outputs of the passive splitters, resulting in the transmission of optical signals over unintended links and hence in higher spectrum occupation with respect to wavelength-switched optical networks (WSONs) based on active devices. To mitigate spectrum waste, FONs can be augmented by deploying programmable optical switches, which increase network flexibility as they allow re-configuration of fiber trees established in FONs to accommodate demands. This filterless network is referred to as Programmable FON (P-FON). In this paper, we propose a traffic-adaptive heuristic algorithm, namely Adapt P-FON, for the re-configuration of programmable optical switches in FONs. The algorithm performs routing and spectrum assignment for traffic demands and also provides optimized configuration of programmable optical switches such that the overall spectrum utilization in the network is minimized. We evaluate the advantages of P-FONs, in terms of spectrum utilization and equipment cost, against FONs and WSON scenarios. Results show that P-FONs have significant advantages in terms of spectrum utilization in comparison to FONs (up to 60%), and, at the same time, cost savings (up to 90%), considering cost of splitters, WSSs and programmable switches, in comparison to WSON.

Content Providers (CPs) typically encrypt the content sent over the telecom network to improve security and privacy of their final users, as well as to protect business-critical information (e.g., contents' popularity). Due to this encryption, Internet Service Providers (ISPs) can not easily apply caching strategies that require the inspection of traffic traversing their networks to select the most popular contents. The most common approach to solve the conflict between privacy and caching consists in allowing a CP to manage the caches (e.g., by storing and delivering the contents) directly from inside the area of the ISP. However, in this way ISPs lose the legitimate control on a portion of traffic traversing their networks. An alternative approach is enabled by recently-proposed architectural solutions that allow a CP to encrypt the contents and associate pseudonyms to them, and the ISP to count the occurrences of such identifiers to infer popularity-related information without inspecting the original contents. However, we observe that ISPs can still obtain valuable information about contents' popularity that may threaten CPs' privacy. In this paper, we formalize a strategy of association between pseudonyms and contents that effectively improves privacy but leads to a degradation of caching performance. We formally define privacy in this context and study the trade-off between caching and privacy considering differerent metrics, such as the hit-rate and the retrieval latency. The results, obtained by means of simulations over both real and synthetic data, show that privacy can be significantly improved while accepting a minor impact on the hit-rate of caching and suggest the applicability of the considered architecture in a real scenario of content delivery.

Traffic demand in fixed and mobile networks is increasing rapidly, driven especially by the growing adoption of Video-on-Demand (VoD) services, which are responsible for roughly 70% of today's Internet's traffic. Network operators must continuously explore new architectural solutions to satisfy increasing traffic at minimum cost. A promising solution consists in deploying caches at the network edge such that VoD requests can be terminated locally. The dimensioning of edge network nodes in terms of storage capacity as well as their placement in the network must be optimized, to reduce costs, improve quality of service, and utilize network resources efficiently. In this paper, we aim to find the optimal deployment of caches, which minimizes overall network resource occupation for VoD service, across the various levels of a hierarchical optical metro network, in terms of the number of caches, their location and dimension (i.e., storage capacity). We develop a discrete-event simulator for dynamic VoD provisioning to measure the performance of different cache deployment strategies in terms of overall network resource occupation and blocking probability. We prove that deploying all the available storage capacity in nearest cache locations does not guarantee the minimal resource occupation. In fact, to minimize resource occupation given a fixed budget in terms of storage capacity, storage capacity must be distributed strategically among caches at different layers of the metro network based on the characteristics of the service, e.g., VoD content catalog popularity distribution.