Distributed training of artificial intelligence models, such as Large Language Models (LLMs), generates highly structured and intense traffic patterns between GPUs, with synchronous and repetitive flows that can easily cause congestion and bottlenecks in data center networks. In this context, currently adopted protocols, such as RoCEv2, show significant limitations in the presence of bursty traffic and low entropy, compromising overall system efficiency. Segment Routing over IPv6 (SRv6) offers a programmable mechanism to steer AI workload traffic along explicitly chosen paths, enabling precise and congestion-aware routing under dynamic conditions. Lightweight monitoring modules can detect congestion conditions in real time and report them to the orchestrator or NICs, enabling dynamic rerouting decisions without requiring control-plane signaling or state in the fabric. SRv6 micro-segment (uSID) encoding allows the NIC to steer traffic along alternate, congestion-free paths simply by updating the IPv6 destination address, preserving RoCEv2 semantics while ensuring rapid adaptability. This work provides a practical implementation and experimental validation of the recent IETF Internet-Draft “SRv6 for Deterministic Path Placement in AI Backends”, demonstrating its feasibility and performance benefits in RoCEv2-based infrastructures. The results highlight the potential of SRv6 as a practical and vendor-agnostic solution to enhance networking efficiency in modern AI datacenters.

Accurate localization is a key enabler for advanced autonomous applications in UAVs and Internet of Things (IoT) networks, particularly in outdoor scenarios where GPS may be unavailable, unreliable, or insufficiently precise. Ultra-Wideband (UWB) technology offers high-accuracy ranging capabilities and has shown strong performance in indoor environments, but its deployment in large-scale outdoor networks remains underexplored. This work investigates the presence of range-dependent bias in long-distance UWB measurements and evaluates its impact on multilateration-based localization systems for UAVs. We present a field measurement campaign with specifically designed UWB testing boards over distances up to 1.2 km, revealing that all tested modules exhibit a growing positive bias, up to 20 cm at the maximum range. To assess how such bias affects positioning, we implement a simulation framework modeling a UWB localization network. We analyze the effect of varying anchor spacing and compare ideal measurements to biased ones using the measured trend. Our results show that geometric anchor deployment can mitigate some of the bias effects, and that awareness of systematic errors is crucial for the design of scalable, accurate outdoor localization networks.

Federated testbeds enable collaborative research by providing access to diverse resources, including computing power, storage, and specialized hardware like GPUs, programmable switches and smart Network Interface Cards (NICs). Efficiently sharing these resources across federated institutions is challenging, particularly when resources are scarce and costly. GPUs are crucial for AI and machine learning research, but their high demand and expense make efficient management essential. Similarly, advanced experimentation on programmable data plane requires very expensive programmable switches (e.g., based on P4) and smart NICs. This paper introduces SHARY (SHaring Any Resource made easY), a dynamic reservation system that simplifies resource booking and management in federated environments. We show that SHARY can be adopted for heterogenous resources, thanks to an adaptation layer tailored for the specific resource considered. Indeed, it can be integrated with FIGO (Federated Infrastructure for GPU Orchestration), which enhances GPU availability through a demand-driven sharing model. By enabling real-time resource sharing and a flexible booking system, FIGO improves access to GPUs, reduces costs, and accelerates research progress. SHARY can be also integrated with SUP4RNET platform to reserve the access of P4 switches.

Nasal polyposis and chronic nasal obstruction significantly impact patients' quality of life and require continuous clinical monitoring. Traditional follow-up methods, based on frequent in-person visits, are often inefficient and burdensome for both patients and healthcare systems. This study presents a telemonitoring solution enabling remote assessment of nasal airflow using a customized version of the Spirobank Smart device. Mechanical and software adaptations were implemented to measure Peak Nasal Inspiratory Flow (PNIF) and transmit data in real time to clinicians via a secure digital infrastructure. The system was validated on 30 healthy volunteers by comparing measurements with those from a standard PNIF Meter. Statistical analysis showed no significant difference between the two devices. The proposed solution offers a reliable, accessible, and costeffective tool for managing chronic nasal conditions remotely, with potential to improve long-term care and patient autonomy.

The monitoring of wildlife through advanced tracking technologies is crucial for ecological research, conservation efforts, and the study of animal behavior. However, most existing tracking devices are primarily designed for positional data collection, with limited integration of biologically relevant sensors. This work presents the design, development, and evaluation of a low-power, multisensor tracking device tailored for long-term wildlife monitoring. Our device integrates a comprehensive set of sensors and novel methodologies to acquire additional information about animals, including monitoring heart rate and respiration in free-ranging individuals. We optimize energy consumption through an advanced power management system, leveraging energy harvesting to ensure long-term operational sustainability. We validate the system with extensive laboratory tests on energy efficiency and data accuracy. This research paves the way for next-generation wildlife tracking devices capable of supporting more comprehensive studies in conservation biology and environmental monitoring.

Recent advances in network switch designs have enabled machine learning inference directly within the switch at line speed. However, hardware constraints limit switches capabilities of tracking stateful features essential for accurate inference, as the demand for these features grows rapidly with line rates. To address this, we propose DIDA, a distributed in-network machine learning approach. In DIDA, feature extraction occurs at the host, features are transmitted via in-band telemetry, and inference is performed on the switches. In this paper, we evaluate the effectiveness and efficiency of this architecture. We examine its impact on network bandwidth, CPU and memory usage at the host, and its robustness across different feature sets and deep neural network classifications.

With the rise of the Network Softwarization era, eBPF has become a hot technology for efficient packet processing on commodity hardware. However the development of custom eBPF solutions is a challenging process that requires highly qualified human resources. Indeed, in eBPF, it is difficult to devise truly modular applications since the development model does not favour the use of pre-compiled functions and libraries. In addition, for safety purposes, each eBFF program must pass a binary code verifier of the Linux kernel, which may increase the difficulty of the development process. To overcome such difficulties and enable a new development model, in this paper we propose the eCLAT framework with the goal to lower the learning curve of engineers by re-using eBPF code in a programmable way. eCLAT offers a high level programming abstraction to eBPF based network programmability, allowing a developer to create custom application logic with no need of understanding the complex details of regular eBPF programming. A developer can write eCLAT scripts in a python-like language to compose eBPF programs. To support such abstraction at the eBPF level, we created an eBPF framework called HIKe which brings code reuse and modularity in eBPF. The eCLAT/HIKe solution does not require any kernel modification. The new development model is tested through two concrete examples and compared with other proposed frameworks in the eBPF world.

With the rise of the Network Softwarization era, eBPF has become a hot technology for efficient packet processing on commodity hardware. However the development of custom eBPF solutions is a challenging process that requires highly qualified human resources. In this paper we propose the eCLAT framework with the goal to lower the learning curve of engineers by re-using eBPF code in a programmable way. eCLAT offers a high level programming abstraction to eBPF based network programmability, allowing a developer to create custom application logic in eBPF with no need of understanding the complex details of regular eBPF programming. To support such modularity at the eBPF level, we created an eBPF library that implements a virtual machine, called HIKe VM. The HIKe VM library extends the conventional eBPF programs so that they can be integrated in eCLAT. The eCLAT/HIKe solution does not require any kernel modification.

Besides providing data sharing, commercial cloud-based storage services (e.g., Dropbox) also enforce access control, i.e. permit users to decide who can access which data. In this paper we advocate the separation between the sharing of data and the access control function. We specifically promote an overlay approach which provides end-to-end encryption and empowers the end users with the possibility to enforce access control policies without involving the cloud provider itself. To this end, our proposal, named ABEBox, relies on the careful combination of i) attribute-based encryption for custom policy definition and management, with ii) proxy re-encryption to provide scalable re-keying and protection to key-scraping attacks, with a novel revocation procedure. Moreover, iii) we concretely embed our protection mechanisms inside a public domain virtual file system module to provide an overlay and trivial-to-use transparent service which can be deployed on top of any arbitrary cloud storage provider.

Segment Routing over IPv6 (SRv6 in short) is a networking architecture suitable for both IP backbones and datacenters. The research, standardization and implementation of this architecture are actively progressing and SRv6 is already adopted in a number of large scale deployments. Effective solutions for Performance Monitoring (PM) of SRv6 networks are strongly needed and there is a lot of activity in this area. A full blown Performance Monitoring solution needs to include: i) Data Plane (as needed to measure metrics such as packet loss and delay); ii) Control Plane (to send commands to the monitoring entities in the nodes); iii) Management Plane (e.g., to collect the measured metrics). Moreover, Big-Data tools and solutions can be applied inside or above the traditional Management Plane boundaries to store and analyze the collected data. In this article we describe SRv6-PM, a solution for Performance Monitoring of SRv6 networks that deals with all the aspects discussed above. SRv6-PM features a cloud-native architecture that supports: i) the ingestion, processing, storage and visualization of PM data using Big-Data tools; ii) the SDN-based control of network routers to drive the performance monitoring operations. In particular, we focus on Loss Monitoring and consider a solution capable of tracking single packet loss events operating in near-real time (e.g., with a time granularity in the order of 10-20 seconds). SRv6-PM is released as open source. We offer a re-usable and extensible platform that can be automatically deployed in different environments, from a single host to multiple servers on private/public clouds and includes a set of Big-Data tools and the SDN control plane. We also provide a reproducible Data Plane environment for PM experiments in SRv6 networks based on the Mininet emulator.

Cryptography key management system plays a very central role in the cloud data security. Nonetheless, a great part of the current commercial solutions rely on cloud providers that hold both the encrypted data and the related private master key of their served customers in their secure key vaults, having a de-facto total control on their customer digital assets. Conversely, entrusting customer companies for key holding can be dangerous as witnessed by many cases of key loss or theft. In this work we present CoProtect, a novel architecture to protect the cryptography keys in cloud systems that leverage on the cooperation between the cloud provider and the customer company. With such trust model, we present the proposed data management strategy, the key generation and the crypto procedures, and a proof of concept.

A careful planning of sleeping activities is essential for guaranteeing long operational time to energy constrained IoT devices, given the extremely low energy consumption required during the sleeping mode. Neighbor discovery instead requires nodes to continuously listen to the radio channel to detect potential and unpredictable messages coming from the arrival of new nodes. These two contrasting needs raise a fundamental question: Should a node be active and listen to the channel or sleep and save energy? Hardware manufacturers introduced dedicated low-power circuitry to allow an energy efficient fast duty cycle (Wake-on Radio) or designed to wake up the main processor when special packets are received (Wake-up Radio). In this work, we propose a novel analytic model optimizing the number of discovered peers for a given energy budget. Our theoretical study demonstrates the existence of an optimum duty cycle in a wide range of operating environments. We designed an adaptive algorithm that dynamically tracks the optimum even in non-stationary conditions, only using local device estimations, therefore allowing nodes to optimize their own duty cycle. Nodes globally converge to a system-wide optimum of a minimal network discovery energy cost.

The expected dramatic growth of connected things raises the issue of how to efficiently organize them, in order to monitor and manage functions and interactions. Information centric networking (ICN) is a communication paradigm that provides content-oriented functionality in the network and at the network level, including content routing, caching, multicast, mobility, data-centric security, and a flexible namespace. Thus, it is a viable solution for supporting Internet of Things (IoT) services without requiring any centralized entity. In this paper, we introduce the lightweight named object solution: a convenient way to represent physical IoT objects in a derived name space, exploiting ICN. We show that this abstraction can: 1) increase the programming simplicity; 2) offer extended functionality, such as augmentation and upgrading, to cope with the “software erosion,” and 3) implement a common interaction logic involving mutual function invocation. We present some proof-of-concept implementations of the proposed abstraction dealing with challenging IoT test cases; we also carry out a performance evaluation in a simulated network scenario.

Many Internet of Things applications require a regular periodic sampling of physical quantities, such as light, CO2, or position. However, for energy harvesting devices, this can be in sharp contrast with the unreliable and time-varying amount of energy gathered opportunistically from the environment, and the severe energy storage limitations in constrained devices further exacerbate such issue. This article proposes an approach devised to jointly optimize frequency and stability of the sampling rate in energy harvesting sensors. Unlike heuristic approaches, StableSENS builds upon a solid theoretical foundation, namely, Lyapunov optimization, which-to the best of our knowledge-is applied here for the first time to the sensing scenario. One of StableSENS' main assets is its very broad applicability: we remark that neither assumptions nor predictions on future energy availability patterns are required. Numerical results obtained in realistic scenarios show that StableSENS yields superior performance with respect to previous heuristic approaches as well as reinforcement-learning-based approaches.

The paper tackles the problem of finding the correct solution to a set of binary choice questions or labeling tasks, by adaptively assigning them to workers in a crowdsourcing system. Such problem becomes quite challenging when we do not initially know neither workers' abilities, nor questions' difficulties (besides common a priori statistics), nor (of course) which is the correct answer. Indeed, such problem requires to jointly learn workers' abilities and questions' difficulties, while adaptively assigning questions to the most appropriate workers so as to maximize our chances to find which are the correct answers. To address such problem, we first cast it into a suitably constructed Bayesian framework which permits us to obtain an analytically tractable (closed form) single-question inference step, and then we address the more general framework via the Expectation Propagation algorithm, an approximated message-passing iterative technique. We then exploit the information gathered by the inference framework as adaptive weights for a maximum weight matching task assignment policy, proposing a computationally efficient algorithm which maximizes the entropy reduction for the questions assigned at each step.

This paper explores methodologies, advantages and challenges related to the use of the Information Centric Network technology for developing NoSQL distributed databases, which are expected to play a central role in the forthcoming IoT and BigData era. ICN services make possible to simplify the development of the database software, improve performance, and provide data-level access control. We use our findings to design a NoSQL spatio-temporal database, named OpenGeoBase, and evaluate its performance by using a real data set relevant to Intelligent Transport System applications.

Neighbor discovery is a primary enabling ability for many emerging mobile applications. Due to its significant impact on the energy budget of battery equipped devices, energy preserving solutions have been investigated for a longtime, often introducing duty cycling. Ultimately, these solutions settle for trade-offs between energy and performance, leaving the final decision on how much energy to allocate to neighbor discovery in the hands of application developers or system engineers. In other words, someone must decide if an improvement in the quality of the discovery (e.g. better discovery latency) is worth an increase in energy consumption. In this paper, we devise a different approach in order to answer the following basic question: how many contacts can a smartphone discover using its battery energy budget? The answer clearly depends on the adopted discovery algorithm, on the mobility conditions, and on the stochastic characteristics of the encounters. However, we demonstrate that there is a natural optimum duty cycle that maximizes the number of discovered contacts in almost every real application scenario. This optimum is natural in the sense that it does not depend on any system level parameter or performance requirements. It depends uniquely on the stochastic characteristics of the meeting process between the nodes. We present an analytic analysis and devise a practical algorithm that dynamically adapts the duty cycle length to the time-varying context, without the need to make assumptions on (or predict) the distribution of the contact duration. The findings presented in the paper are validated against data coming from real human mobility traces and implemented on a mobile application.

Algorithms and protocols for opportunistic, delay tolerant and wireless ad-hoc networks are designed and validated by simulating the people interactions induced by the nodes mobility. There are cases in which we are interested in simulating just the first interaction between a pair of nodes, for instance to assess the performance of a discovery or epidemic routing protocol. In this cases nodes rapidly extinguish their utility hence it is not convenient to simulate these scenarios using a fixed amount of nodes. Thus we present a novel simulation methodology that introduces the "open environment" where nodes enter, can interact through meeting with other nodes and then exit, keeping the focus on the environment rather than on what happen before and after the nodes stay in the interesting area. The proposed approach uses the statistical distribution extracted from the real traces to reproduce directly the human interaction pattern without going through the traditional random way point approach. Meetings are simulated by a time-varying graph that holds the state of the interactions in the environment, while adapting to the statistics of single node to its history. We show that even in a simple scenario, the epidemic infection, Markov memory-less models have been fairly far from the interaction scenarios that the method reproduces.

The aim of this extended abstract is to present Campus++, a location-based publish-subscribe system for intermittently connected delay tolerant networks, exploiting IEEE 802.15.4 devices, and taking into due account the severe constraints deriving from their physical characteristics. We describe our proposed architectural model and how we implemented our solution in a real test-bed. We point out that our system can be easily adapted to operate in a fully distributed, infrastructure-less way, allowing free communications e.g. in disaster areas or in areas in which ”usual” communications means are either non existent or intentionally made unavailable.

In this paper we present a Robust data Replication Algorithm (RRA) for MANETs. RRA ensures persistent data availability by duplicating data over replica-nodes, while minimizing the consumption of radio resources. RRA is the first algorithm thoroughly designed to operate in environments with node failures and radio obstacles (e.g., buildings, walls). Our challenge is to consider such impairments as "accidental"; i.e., their occurrence neither can be run-time predicted, nor can be a-priori quantified in terms of probability. As instance, this is the case of nodes wandering in an unknown territory. To cope efficiently with the novel challenge, we suggest a practical assumption: during a refresh period, only a single accidental impairment may occur. Following this assumption, we succeed in formalizing (and solving) the resource minimization problem as the finding of a minimum-cost-set-cover. We assess RRA performance through NS2 simulations. Simulation code is released as open-source.

In this paper, we propose and evaluate an overlay distribution algorithm for P2P, chunk-based, streaming systems over forest-based topologies. In such systems, the stream is divided in chunks; chunks are delivered by each node in a store-and-forward way. A relaying node starts distributing a chunk only when it has completed its reception from another node. Peers are logically organized in a forest of trees, where each tree includes all peers. The source periodically distributes different chunks to each tree for their delivery. Our key idea consists in employing serial transmission: for each tree, and thus, for each chunk, the source node sends the chunk to its children in series; the same holds for each peer node of the tree, excluding the leaves. Besides this basic idea, the contributions of this paper are: 1) we demonstrate the feasibility of serial transmission over a forest of trees, which is not a trivial problem, unlike the case of parallel transmission; 2) we derive an analytical model to evaluate the system performance; 3) we derive a theoretical bound for the number of nodes reachable in a given time interval or equivalently for the time required to reach a given number of nodes; 4) we prove the optimality of our approach in terms of its capability to reach such bound; and 5) we develop a general simulation package for P2P streaming systems and use it to compare our solution to literature results. Finally, we stress that this paper is focused on the theoretical properties and performance understanding of the proposed distribution algorithm, rather than on its practical implementation in a real system. However, we also briefly describe a practical workable implementation of our algorithm.

Many distribution algorithms have been proposed up to now for P2P real time streaming. However, due to the lack of basic theoretical results and bounds, common sense and intuitions and heuristics have driven their design so far. The consequence is that we can find in the literature a large variety of different choices about the main aspects of a P2P system, such as overlay topology, scheduling process and upload strategy. In this situation, it is difficult to establish unambiguously the absolute goodness of a particular algorithm or even the rationale behind a particular choice or solution. In this paper we propose and evaluate a theory-driven distribution algorithm for P2P real time streaming. We take advantage from a previous theoretical study, where: i) we derived a theoretical performance bound for forest-based overlay topologies regarding the number of nodes reachable in a given time interval or equivalently the time required to reach a given number of nodes; ii) we proved the optimality of streamline, a distribution algorithm based on the serial transmission over forest-based topologies, in terms of its capability to reach such a bound. The streamline algorithm is based on some ideal assumptions that prevent its practical implementation. In this paper we remove these assumptions and present a practical and working algorithm, named operational streamline or simply O-Streamline. We also evaluate the performance of O-Streamline, comparing them with the optimal bounds of streamline.