In current computing processing infrastructures, optimizing performance and energy efficiency across heterogeneous hardware environments remains a critical challenge. Recently, the growth of different types of hardware architectures has raised challenges but also opportunities. In this paper, we present Decomposer, designed to decompose monolithic applications into modular components, allowing them to be executed across various combinations of hardware architectures, both locally and in distributed settings. Using static and dynamic analysis, Decomposer allows flexible allocation of functions to different targets, optimizing both execution time and energy consumption. Our approach involves the generation of Intermediate Representations (IR) using LLVM and MLIR, allowing the identification of functions for decomposition and their distributed execution on heterogeneous hardware. By expanding execution options, Decomposer provides flexibility in resource allocation without increasing execution times. Furthermore, we observed that such decomposition improves cache utilization, reducing L1 cache misses and L2 references in more than 50%, thus enhancing memory efficiency. This work, although in its initial phase, shows that this approach allows for dynamic workload distribution across multiple targets, offering a more versatile and adaptable solution for application execution, aligning resource usage with operational objectives and environmental considerations. We envisage that the compilers area and its techniques can be very useful for optimizing a set of applications to be decomposed and run not only in core data centers but also at the edge 5G network.
According to the World Health Organization, there were 1.19 million fatalities in road accidents in the year 2023 with more than half being Vulnerable Road Users (VRUs) — while modern vehicles have advanced safety features, pedestrians, bicyclists, and motorcyclists often lack protection. This paper discusses the possibilities of incorporating Mixed Reality (MR), and V2X communications within smart cities to create emerging ITS applications focused on VRUs. In addition, we discuss the architecture of smart cities and Multi-access Edge Computing (MEC) that are aimed to decrease latency and enhance data processing. Our review also highlights major challenges such as user acceptance, scalability, and the real-world applicability of the technology in living labs. As a result, we propose areas for future research that incorporate MR interfaces, geographically distributed edge systems, and novel interaction techniques with autonomous traffic and vehicles. This paper explores new pathways for urban mobility, striving for a safer and more equitable environment for all road users in cities.
We present a prototype that could open the doors to mitigate digital exclusion in technologically underserved pop-ulations. By converging software-defined networks (SDN) with programmable data planes (PDP), our approach enables one to customize low-cost hardware to fit the network behavior. We integrate low-power wide-area networks (LPWANs), known for their scalability and efficiency, with PDPs to propose a flexible solution for changing requirements, such as distance and terrain configurations, validated through real-world test scenarios.
This paper introduces the Encrypted Network Traffic Analysis (ENTA) platform, a scalable AI-driven system de-signed for traffic analysis with support for identifying encrypted VoIP traffic generated by instant messaging applications (IMAs). User behaviors, such as exchanging audio messages through different IMAs, are emulated, and the resulting network traffic is captured for analysis. The ENTA platform's capabilities are demonstrated in feature extraction, data pre-processing, AI model training and testing, and generating key performance indi-cators (KPIs) to assist network operations teams. This demonstration showcases the ENTA system's end-to-end functionality with a particular emphasis on accurately classifying and identifying IMA VoIP traffic.
When sending flows to arbitrary destinations, current multihoming routers adopt simple congestion-oblivious mechanisms. Therefore, they cannot avoid congested paths. In this paper, we introduce 2SYN, the first congestion-aware multihoming algorithm that works for any destination. We explain how it dynamically selects a preferred path for new connections, even given previously-unseen destinations. We further demonstrate that it can be easily implemented in Linux. Finally, in a real-world experiment with either LTE or a wired link, we show how 2SYN dynamically adapts to the quality of the connection and outperforms alternative approaches. Thus, 2SYN helps companies better manage their networks by leveraging their multihoming capabilities.
The Border Gateway Protocol (BGP) enables communication between Autonomous Systems (ASes) on the Internet. BGP offers significant flexibility for traffic engineering through BGP communities, which are operatordefined tags that convey information or requests in route announcements. Unfortunately, the absence of standardized semantics or centralized repositories for BGP communities complicates and limits their use, hindering the effective management of interdomain routing. This thesis develops techniques to infer BGP community semantics using public BGP data from routing collectors, overcoming the lack of documentation and providing datasets that can be automatically updated. We first propose a set of techniques to infer location communities, which are communities related to entities or locations traversed by a route. We apply our techniques to billions of routing records from public BGP collectors and show that they produce high precision (ranging from 86% to 93%) and recall (ranging from 72% to 81%). We also design and evaluate algorithms to automatically uncover BGP action communities and ASes that violate standard practices, revealing undocumented relationships between them (e.g., sibling relationships). Our experimental evaluation uncovers previously unknown AS relationships and shows that our algorithm to identify action communities achieves average precision and recall of 92.5% and 86.5%, respectively.
5G technology addresses user privacy concerns in cellular networking by encrypting subscriber identifier with ellip-tic curve-based encryption and then transmitting it as ciphertext known as Subscriber Concealed Identifier (SUCI). However, an adversary equipped with a quantum computer can break a discrete logarithm-based elliptic curve algorithm and the user privacy in 5G is at stake against quantum attacks. In this paper, we study the incorporation of the post-quantum ciphers in the SUCI calculation both at the user equipment and at the core network, which involves the shared key exchange and then using the resulting key for the ID encryption. We experiment on different hardware platforms to analyze the PQC key exchange and encryption using NIST-standardized CRYSTALS-Kyber. Our analyses focus on the performances and compare the Kyber-based key exchange and encryption with the current (pre-quantum) Elliptic-curve Diffie-Hellman (ECDH). The performance analyses are critical because mobile networking involves resource-limited and battery-operating mobile devices. We measure and analyze not only the time and CPU-processing performances but also the energy and power performances. Our results show that Kyber-512 is the most efficient and even has better performance (i.e., faster computations, and lower energy consumption) than ECDH.
In today's fast-changing industrial landscape, organizations need resilience at both individual and organizational levels to stay competitive. The XPERTISE 5.0 research agenda proposes a deep-tech approach to enhancing workforce adaptability through artificial intelligence (AI), extended reality (XR), brain-computer interfaces (BCI), and adaptive learning systems. It examines how digital learning environments can foster resilience by offering personalized and tech-driven trainings. The research aims to develop a modular learning framework for organizations to customize training programs based on their needs, technology, and workforce capabilities. Using AI-driven personalization, real-time physiological feedback, and immersive simulations, this approach enhances knowledge retention, decision-making, and problem-solving skills. It also emphasizes human-centric digital transformation to help employees navigate technological disruptions. XPERTISE 5.0 supports the Industry 5.0 paradigm, highlighting resilience, sustainability, and human-technology collaboration. This research lays the foundation for scalable and adaptive training models across industries, promoting a future-ready workforce that embraces technological change and enhances organizational agility and innovation.
Network datasets are essential resources to evaluate the effectiveness of new technologies in diverse network environments. Unfortunately, these datasets are often not publicly available or lack the necessary density and diversity for thorough testing. Generative Adversarial Networks (GANs) have shown promise in generating realistic synthetic data. Transfer Learning (TL), which transfers knowledge from one domain to another, is another technique that potentially enhances data generation. This paper combines such techniques in a novel framework capable of generating synthetic network data tailored to specific desired protocol types. A GAN is pre-trained on an available dataset with substantial size and diversity corresponding to a source protocol. Then, exploring a TL method, the GAN undergoes fine-tuning using a smaller target protocol dataset. This fine-tuning allows the GAN to generate an augmented dataset containing relevant samples from the target domain. Developed experiments include two different data generation scenarios: i) Intra-protocol: transferring knowledge from datasets with different characteristics, considering the same source and target protocols; ii) Inter-protocol: transferring knowledge between protocols from different layers of the ISO-OSI reference model. The obtained results demonstrate that the proposed GAN and TL framework effectively generates high-quality synthetic traffic, featuring a significant rate of well-formed packets; a high percentage of packets with an appropriate query response; and a considerable similarity between the generated and real packets - measured by the FID (Frechet Inception Distance) - compared to standard GANs without fine-tuning.
High-quality communication technology is essential in our globalized, digital world. However, network issues and poor mobile connectivity still impact service quality and user experience across various applications. The roll-out of 5G networks promised improvements in network quality including lower latency and better throughput. But questions whether these improvements have materialized, about regional differences in mobile network quality, and whether the current network infrastructure is prepared to support today's, currently emerging, and future applications in a high quality remain. To address these questions, we analyzed over 225 million network quality measurements from 2022 and 2023 provided by the company OpenSignal to assess mobile network quality as perceived by an end user along all roadways in the German state Bavaria. By examining network generations, throughput, and latency metrics, we provide valuable insights for network planning, helping both engineers and the public in understanding the current state of mobile connectivity on Germany's roads.
The rapid adoption of containerized cloud environments requires robust and efficient Auto-Scaling (AS) mechanisms to ensure adequate resource utilization, high performance, and cost-effectiveness. Traditional AS approaches, often based on predefined thresholds, fail to adapt well to dynamic workloads. This paper investigates the potential of Reinforcement Learning (RL) as a generalized solution for efficient AS in containerized clouds. Building on previous studies, this paper examines whether RL approaches can learn adaptive scaling policies when trained on diverse workload datasets and tested across different scenarios. A Multi-Objective (MO) reward function has been designed to optimize key performance factors such as the application's response time, and resource utilization. The results demonstrate that RL algorithms can effectively balance competing objectives and adapt to changing workloads. The Latency strategy resulted in lower latency but required more pods (7.4) and slightly higher CPU usage (28.92%). In contrast, the Cost strategy minimized deployment costs with fewer pods (3.56) and lower CPU usage (24.45%). This study highlights the versatility and efficiency of RL in managing complex, real-time scaling decisions in containerized cloud infrastructures.
Deep Neural Networks (DNNs) are complex and versatile Machine Learning (ML) algorithms, essential to many systems. A single system can run multiple DNNs simultaneously, each handling different tasks to meet diverse user requests. Organizational DNN deployments must adhere to strict response time targets defined by Service Level Agreements (SLAs), as failure to meet these targets can incur significant costs. Therefore, accurately predicting performance under various workloads is crucial for optimizing resource allocation and avoiding SLA violations. Despite much research on DNNs, there is limited literature on predicting their performance across varying workload-resource configurations, with existing studies often neglecting important variables or lacking accuracy. Furthermore, predicting the performance of co-hosted DNNs that share resources is an understudied area. This paper addresses this gap by examining the response time behavior of co-hosted DNNs under various resource, workload, and DNN-related factors. We propose an ML-based performance modeling strategy to predict the likelihood of meeting predefined SLA-driven response time targets in co-hosted deployments. Our study identifies key factors related to the host system and the DNNs' isolated performance, enabling effective prediction without extensive historical data collection. This approach helps mitigate potential SLA violations proactively. Our model achieves an SLA compliance prediction accuracy of 90.2% and an f-score of 98.0%, demonstrating strong generalization to unseen workloads and resource configurations.
Autonomous Network enablers for a 5G Network-on-Wheels are investigated, with a field-trial for a Communication Service Provider. The objective was to devise solutions to reduce time spent on network life cycle tasks, such as network function validation, system commission and service deployment. The field trial revealed significant data collection and analysis challenges even with a single cell campaign, this suggests severe costs with public-network scales. The work contributes a new architecture for experiments and optimisation of observability strategies with algorithms in 5G networks. This work helps operators more effectively deploy 5G networks and services.
Numerous research contributions and monitoring systems in Interdomain Routing rely on data captured from specific vantage points on the Internet, referred to as Collector Peers, and collected by entities known as Route Collectors. In this paper, we first argue that current methods used to determine whether Collector Peers and Route Collectors are functioning correctly within a specific time frame are unreliable. To address this, we propose a method to assess the accuracy of the existing measurement framework. Our approach focuses on identifying sequences of update messages within Interdomain signaling, analyzing their start and end times, as well as their frequencies. We validate the accuracy of our method through two steps: (1) Analyzing sequences generated by a set of 'beacons' that emit Interdomain signaling at various frequencies, in order to evaluate our ability to characterize these sequence frequencies. (2) Assessing our ability to identify malfunctioning Collector Peers by comparing our results with known faults, which serve as the ground truth.
The effectiveness of cybersecurity research for SCADA systems depends on access to high-quality network traffic data, yet such data remains scarce due to proprietary restrictions and security concerns. Synthetic data generated by machine learning models, particularly Generative Adversarial Networks (GANs), presents a promising alternative. This study provides a preliminary evaluation of GAN-based approaches for SCADA network traffic synthesis using the IEEE ITACHA DNP3 Smart Grid dataset. We compare a general-purpose GAN (CTGAN) with a network-traffic-specific GAN (NetShare) based on fidelity and statistical consistency. Initial results indicate that CTGAN generates statistically diverse synthetic data, while NetShare suffers from excessive duplication, limiting its applicability. These findings offer an early structured roadmap for selecting and refining generative models for SCADA data synthesis, supporting future research in smart grid security.
The performance benefits of data plane programmability have motivated many researchers to offload the computation of applications that previously operated only on servers to the network, creating the notion of in-network computing (INC). Because failures can occur in the data plane, fault tolerance mechanisms are essential for INC. However, INC operators and developers must manually set fault tolerance requirements using domain knowledge to change the source code. These manually set requirements may take time and lead to errors in case of misconfiguration. In this work, we present ARAUCARIA, a system that composes fault tolerance building blocks for INC based on high-level intents. The system allows the specification of requirements using an intent language, which allows the expression of consistency and availability requirements in a constrained natural language. A refinement process translates the intent and instruments the INC with essential building blocks and configurations. Our prototype of ARAUCARIA enables fault tolerance for INC applications on BMv2 and in a testbed with Tofino ASICs. Experiments show that the system provides fault tolerance with negligible overhead.
This paper presents the methodology for using LLMs to ease core network signaling analysis. This is done by applying RAG techniques to process standards that describe protocol data formats - and then asking natural language questions about actual capture traces. Analyzing 5G networks is very challenging due to the complex and dynamic nature of signaling protocols. Unlike previous generations, protocol fields and values are described in a human-readable format, enabling textual post-processing, and the direct application of LLM models. Intent-based network management involves natural language-based human interaction with the networking equipment so the desired outcome is achieved without step-by-step instructions and settings by the human. This paper proposes a novel approach that uses the combination of Retrieval Augmented Generation (RAG) and Langchain to automatically answer human questions regarding signalling data. This toolchain makes the analysis part of network fault management intent-based. Moreover, by training LLMs on a vast corpus of standardized signaling data, we demonstrate the model's ability to generate realistic test data. This approach improves the efficiency of automated test environments, ensuring the reliability and performance of networks in real-world conditions.
The complexity of modern networks, combined with business requirements and human oversight, presents significant management challenges. This thesis focuses on developing a holistic management system, Emergence, which integrates intent-based networking, large language models (LLMs), and policy-driven automation through closed control loops. Our contributions are: 1) Formalizing intents into a hierarchy of policies at various abstraction levels; 2) Intelligent and automated intent-to-policy decomposition using generic pre-trained LLMs, resulting in a Policy Tree—an ordered set of Monitor-Analyze-Plan-Execute (MAPE-K) policies; 3) Automated and scalable intent deployment through control loops and Finite State Machines for policy execution; 4) Monitoring and mitigating intent drift using LLMs and additional tools to assure intents in response to changing conditions. Our solution provides a robust, scalable, and intelligent system for modern network management, fulfilling and assuring intents 1-3x faster on average compared to manual procedures. Moreover, the policy-based approach enhances explainability and control over management decisions and actions. Lastly, we share future directions towards trustworthiness.
Network Slice Identification (NSI) is crucial for managing Quality of Service (QoS) in Beyond 5G (B5G) networks, particularly for smart city applications. This study explores and compares supervised, unsupervised, and semi-supervised learning techniques for NSI, addressing the challenges of limited labelled data in production environments. We use a publicly available 5G network dataset to model and perform comparisons among supervised, unsupervised, and semi-supervised learning approaches. Our methodology involves feature selection, dimensionality reduction using t-SNE, and addressing class imbalance through undersampling. We evaluate model performance using accuracy and Silhouette Score. Our results show a Random Forest Classifier achieves 100% accuracy with supervised learning. The unsupervised K-Means clustering model, optimized with both t-SNE and undersampling, achieves a mean accuracy of 92.83%. Semi-supervised learning using a self-training method and being trained on only 10% of the training data points performs comparably to the supervised models. Importantly, we demonstrate the robustness check using test data perturbation injecting additional variability in data simulating unknown 5G network fluctuations. This comparative analysis provides insights into the trade-offs between different learning approaches for NSI in B5G networks, offering practical solutions for scenarios with different conditions of labelled data.
The digital divide remains one of the biggest challenges of the modern era, limiting access to information and global connectivity for populations in remote or rural regions. The doctoral research goes in the direction of an integrated approach to address the divide in the view of 6G. In particular, the integration of Non-Terrestrial Networks (NTN) and Terrestrial Networks (TN) is envisioned as a key part of the future 6G systems. Exploiting the Software-Defined Wide Area Network (SD-WAN) technology, we aim to propose an architecture for an adaptive and resilient network that can integrate satellite technologies such as Low Earth Orbit (LEO) satellite networks. Moreover, we present some results of using Reinforcement Learning (RL) in the tunnel selection problem for SD-WAN. In order to assess the performance of the LEO satellite, we are also collecting and analyzing real measurements using a Starlink connection. Finally, we propose some open questions in the field of integrating NTN/TN with SD-WAN.
