Over the last couple of years, large language models have increasingly been integrated into many computing applications. For privacy preservation, they are now deployed on edge devices. However, these deployments are vulnerable to bit flip attacks and backdoor attacks that compromise the integrity of the model. Traditional remote attestation techniques fail to detect such manipulations due to the large model size and the stealthiness of the attacks.In this paper, we present TrustSeed, a lightweight functional attestation protocol that uses a single inference to ensure large language models’ integrity. TrustSeed verifies integrity by applying deterministic, seed-based modifications to model weights within a Trusted Execution Environment and comparing the last intermediate activations and output distribution against a golden reference on the verifier. This approach prevents precomputed or forged responses, ensuring freshness and unpredictability in each attestation round. Our analysis shows that output distribution and last intermediate activations are effective indicators of integrity. We test TrustSeed against bit-flip, data poisoning, and weight poisoning attacks, reliably detecting even single-bit alterations. Extensive evaluations on edge platforms and an HPC system demonstrate minimal overhead and up to 127× faster attestation compared to state-of-the-art full-model hashing.

AI-based mission-critical software exposes a blessing and a curse: its inherent statistical nature allows for flexibility in result quality, yet the mission-critical importance demands adherence to stringent constraints such as execution deadlines. This creates a space for trade-offs between the Quality of Result (QoR)-a metric that quantifies the quality of a computational outcome-and other application attributes like execution time and energy, particularly in real-time scenarios. Fluctuating resource constraints, such as data transfer to a remote server over unstable network connections, are prevalent in mobile and edge computing environments-encompassing use cases like Vehicle-to-Everything, drone swarms, or social-VR scenarios. We introduce a novel approach that enables software engineers to easily specify alternative AI service chains-sequences of AI services encapsulated in microservices aiming to achieve a predefined goal-with varying QoR and resource requirements. Our methodology facilitates dynamic optimization at runtime, which is automatically driven by the MARQ framework. Our evaluations show that MARQ can be used effectively for the dynamic selection of AI service chains in real-time while maintaining the required application constraints of mission-critical AI software. Notably, our approach achieves a 100x acceleration in service chain selection and an average 10% improvement in QoR compared to existing methods.

Physical Unclonable Functions (PUFs) serve as a lightweight source to generate cryptographic keys utilizing the inherent physical device properties, making them particularly suitable for resource-constrained environments such as Internet of Things (IoT) devices. Recently, Physical-Related Functions (PReFs) extended PUFs to enable multiple devices to generate similar keys without the need to exchange or store them, improving security. However, state-of-the-art PReF implementations rely on a Trusted Third Party (TTP) to identify relative challenges, introducing a potential vulnerability if the TTP is compromised. In this work, we propose the first decentralized PReF protocol, removing reliance on the TTP and mitigating associated security risks. The proposed protocol allows relative challenges to be identified directly between devices in a decentralized manner. Additionally, we formalize a mathematical model to estimate the minimum number of devices required to build a network, based on the sizes of the PUF and the shared Challenge-Response Pair (CRP).. We demonstrate the generality of our model by verifying it across different types of state-of-the-art PUFs (Arbiter-based Non-Volatile Memory PUF (ANV-PUF) and Pseudo Linear Feedback Shift Register PUF (PLPUF).). We establish a 128 bit cryptographic key using the proposed protocol that matches the state-of-the-art but in a decentralized manner. Moreover, we prove that our protocol can be used to construct hardware-assisted attestation networks using ANV-PUF and PLPUF implementations with a shared secret of 16 bit that allows for both integrity and identity verification.

Federated Learning (FL) is an emerging machine learning paradigm that enables the collaborative training of a shared global model across distributed clients while keeping the data decentralized. Recent works on designing systems for efficient FL have shown that utilizing serverless computing technologies, particularly Function-as-a-Service (FaaS) for FL, can enhance resource efficiency, reduce training costs, and alleviate the complex infrastructure management burden on data holders. However, current serverless FL systems still suffer from the presence of stragglers, i.e., slow clients that impede the collaborative training process. While strategies aimed at mitigating stragglers in these systems have been proposed, they overlook the diverse hardware resource configurations among FL clients. To this end, we present Apodotiko, a novel asynchronous training strategy designed for serverless FL. Our strategy incorporates a scoring mechanism that evaluates each client’s hardware capacity and dataset size to intelligently prioritize and select clients for each training round, thereby minimizing the effects of stragglers on system performance. We comprehensively evaluate Apodotiko across diverse datasets, considering a mix of CPU and GPU clients, and compare its performance against five other FL training strategies. Results from our experiments demonstrate that Apodotiko outperforms other FL training strategies, achieving an average speedup of 2.75x and a maximum speedup of 7.03x. Furthermore, our strategy significantly reduces cold starts by a factor of four on average, demonstrating suitability in serverless environments.

While the concept of AI4Net has been widely discussed in the past decade and adopted in 5G, its counterpart, Net4AI, has not gained that much attention so far. This is mostly due to the absence of solutions for the network to support AI applications beyond providing the communication infrastructure. In-Network Computing (INC) is a promising paradigm, potentially being integrated into 6G, which opens new solutions for realizing Net4AI. This paper focuses on the specific case of INC-assisted Split-AI. A Neural Network (NN) is split vertically and the executions of some layers of the split NN are offloaded to the entities of an Intelligent 6G User Plane. With the example of INC-assisted Split-AI, we elaborate on the challenges of Net4AI and discuss key requirements for the 6G architecture in terms of novel capabilities and information exchange.

Edge computing brings added benefits for different elements in the overall system (e.g., users, operators and service providers). However, currently there are no proper interfaces and mechanisms to instantiate third-party services within the operators' infrastructure (e.g., as a MEC application), thus hindering edge computing to reach its full potential. To fill this gap, this paper presents architectural enhancements, interfaces and mechanisms to enable dynamic and efficient third-party service deployment within the operators' domain. A simulation-based analysis is presented to showcase the relevance of the proposed solution. Results highlighted the benefits of optimal migration of third-party services into a distributed setting, compared to the unveiled drawbacks of a centralized approach. In addition, the key components of the solution are implemented and experimentally validated through a proof-of-concept prototype showcasing the performance impact of the proposed approach as well as the suitability of its implementation.

Multiaccess edge computing (MEC) service migration is a technology whose key objective is to support ultralow-latency access to services. However, the complex ultralarge-scale edge service migration problem requires extensive research efforts, regarding the foreseen ultradensified edge nodes in 5G and beyond. In this article, we propose a novel dynamic service migration optimization architecture for ultralarge-scale MEC networks. We develop a new multicriteria decision-making algorithm: Technique for order of preference by similarity to ideal solution with attribute-based Niche count, named TOPANSIS, which showcases its strength to provide an optimal solution for service migration in large-scale deployments towards optimal data rate, latency, and load balancing. We further decentralize the operation of TOPANSIS to release the traffic burden from central datacenters by leveraging local decision making by edge nodes, while relying on central cloud coordination to account for the overall network information. Simulation results showcase that the proposed architecture outperforms the selected benchmarks with an average improvement of 39.41% for latency, 2.92% for data rate, as well as 10.53% and 6.26% for RAM and CPU load balancing, respectively. Moreover, the feasibility of the proposed solution is validated by means of a proof-of-concept implementation and experimental assessments.

Federated Learning (FL) is a machine learning paradigm that enables the training of a shared global model across distributed clients while keeping the training data local. While most prior work on designing systems for FL has focused on using stateful always running components, recent work has shown that components in an FL system can greatly benefit from the usage of serverless computing and Function-as-a-Service technologies. To this end, distributed training of models with severless FL systems can be more resource-efficient and cheaper than conventional FL systems. However, serverless FL systems still suffer from the presence of stragglers, i.e., slow clients due to their resource and statistical heterogeneity. While several strategies have been proposed for mitigating stragglers in FL, most methodologies do not account for the particular characteristics of serverless environments, i.e., cold-starts, performance variations, and the ephemeral stateless nature of the function instances. Towards this, we propose FedLesScan, a novel clustering-based semi-asynchronous training strategy, specifically tailored for serverless F L. FedLesScan dynamically adapts to the behavior of clients and minimizes the effect of stragglers on the overall system. We implement our strategy by extending an open-source serverless FL system called FedLess. Moreover, we comprehensively evaluate our strategy using the 2ndgeneration Google Cloud Functions with four datasets and varying percentages of stragglers. Results from our experiments show that compared to other approaches FedLesScan reduces training time and cost by an average of 8% and 20% respectively while utilizing clients better with an average increase in the effective update ratio of 17.75%.

In this paper, we modeled the emerging edge-computing-enabled live service migration as a multi-criteria problem optimization, tackling migration costs and benefits, as well as discussion of service providers' data privacy, simultaneously. Based on the optimization formulation, we conducted a small-scale analytical feasibility test, considering widely-utilized multi-criteria decision making algorithms, based on which we proposed a new TOPSIS based service migration algorithm. The algorithm was evaluated using simulations, whose results show that the proposed algorithm is sufficient to support live service migration for heterogeneous edge computing, while outperforming benchmarks in with respect to reducing migration costs and increasing achieved benefits, by 34.52% and 60.21%, respectively.

The proximity of Mobile Edge Computing offers the potential for offloading low latency closed-loop applications from mobile devices. However, to repair decreases in quality of service (QoS), e.g., resulting from user mobility, the placement of service instances must be continually updated - essential for mission critical applications that cannot tolerate decreased QoS, for example virtual reality or networked control systems. This paper presents BigMEC, a decentralized service placement algorithm that achieves scalable, fast, and high-quality placements by making local service migration decisions immediately when a drop in QoS is detected. The algorithm relies on reinforcement learning to adapt to unknown scenarios and to approximate long-term optimal placement updates by taking future transition costs into account. BigMEC limits each decentralized migration decision to nearby edge sites. Thus, decision computation times are independent of the number of nodes in the network and well below 10ms in our experimental setup. Our ablation study validates that, using its scalable approach to decentralized resource conflict resolution, BigMEC quickly approaches optimal placement with increasing local view size, and that it can reliably learn to approximate long-term optimal migration decisions, given only a black-box optimization objective.

Hardware acceleration of network functions is essential to meet the challenging Quality of Service requirements in nowadays computer networks. Graphical Processing Units (GPU) are a widely deployed technology that can also be used for computing tasks, including acceleration of network functions. In this work, we demonstrate how commodity GPUs, which do not provide any network interfaces, can be used to accelerate network functions. Our approach leverages PCIe peer-to-peer capabilities and allows the GPU to control the network interface card directly, without any assistance from the operating system or control application. The presented evaluation results demonstrate the feasibility of our approach and its performance of up to 10 Gbit/s, even for small packets.

Peer-assisted delivery of video content has shown a great potential to reduce upload bandwidth requirements for content providers by exploiting idle client resources in the video dissemination process. As primary content sources, the servers run by content providers play a critical role in such systems, making their adequate provisioning a key part of the streaming mechanism. While dynamic resource provisioning has been studied before, little is known about resource allocation for streaming of scalable media content. Besides the pure amount of resources, here, the quality level of the delivered video content becomes relevant. The spreading of video blocks with the wrong quality can lead to situations where peers are forced to reduce their video qualities, despite them having enough download capacity. To address this problem, in this paper, a new SVC-based adaptation policy and a request-based extension to it are proposed, enabling content providers to manage their streaming services in a video quality-aware manner. Prototypical evaluations show that the mechanisms outperform existing quality-agnostic approaches in terms of delivered SVC video quality.

P2P Video-on-Demand (VoD) based on Scalable Video Coding (SVC) (the scalable extension of the H.264/AVC standard) is gaining momentum in the research community, as it provides elegant adaptation to heterogeneous resources and network dynamics. The major question is, how do the adaptation algorithms and designs affect the overall perceived performance of the system? Better yet, how can the performance of an SVC-based VoD system be defined? This paper explores the impact and trade-offs of SVC-based quality adaptation with focus on the SVC layer selection algorithms, which are performed at different streaming stages. We carry out extensive experiments to evaluate the performance in terms of session quality (start-up delay, video stalls) and delivered SVC video quality (layer switches, received layers), and find out that these two metrics exhibit a trade-off. Our analysis and conclusions give multimedia providers insights on how to design and fine-tune their VoD system in order to achieve best performance.

There are currently two concurrent trends in the Internet. First, the number of Internet users and their connection speeds are increasing rapidly. Second, Internet-based applications are dominating how people receive information, communicate, and entertain themselves. Therefore, we are witnessing an enormous increase in IP-based multimedia traffic, which is putting an enormous strain on the network. Additionally, router and network virtualization are gaining importance, enabling more intelligent networks. Therefore, we argue that networks should not be merely bystanders to this multimedia revolution. In this paper we present a media-aware network solution based on router virtualization that aims at striking a balance between intelligence and adaptation at the edge and in the core of the network. Using an extensive simulative study, we demonstrate that our media-aware network not only helps in enhancing streaming performance during bottlenecks, but also minimizes the side effects of congestions on user perceived quality, making it a need for future Internet multimedia applications.

Energy consumption is responsible for a large fraction of costs in today's content distribution networks. In upcoming decentralized architectures based on set-top boxes (STB), acting as tiny servers, idle times can dominate distribution costs, since no cooling costs occurs and the Internet access is often paid in a flat-rate manner. The often assumed always-on property of STBs provides high availability but might also waste up to 93% of the baseline energy. In this paper we consider suitable standby policies that reduce energy consumption but still allow offloading content servers significantly. We devise optimal and heuristic standby policies and evaluate them in a realistic scenario to show that a near-optimal behavior can be reached by utilizing the specific features of STBs.

P2P streaming has attracted much attention recently with promises for higher revenues and better load distribution. Still, the majority of P2P video streaming systems today employ the one-size-fits-all concept where the same video bit-rate is offered to all users. Here the promising H.264/Scalable Video Coding (SVC) standard is seen as a necessity in not only supporting heterogeneous resources, but also in reducing the impact of P2P dynamics on the perceived Quality-of-Experience (QoE). In this demonstration we present our streaming application that uses SVC to adapt to different user requirements and resources. The application employs a novel QoE- aware layer selection algorithm that maximizes flexibility through SVC while taking impact on QoE into consideration.

P2P Streaming has attracted much attention recently with promises for more revenues and better load distribution. In parallel, social networking has changed how people interact using the web. One interesting use-case for next generation IPTV is Social TV. In such a system, users are able to watch some media stream and interact with each other at the same time. While deploying P2P Social TV, one inherent problem in P2P streaming systems remains, how to incite users to contribute. In this demonstration we show how social networks can be used to build new incentive mechanisms. Rather than making social relations a mere addition, we build our streaming system on top of a user's social network. This design greatly simplifies the system and requires no further entities for management. In this demonstration we present the first version of StreamSocial that, based on a plug-in based design, allows users to stream videos while performing social interactions.

Peer-to-Peer (P2P) systems gained popularity and are responsible for a large share of today's Internet traffic. Nevertheless, their dynamic nature and the intended lack of control through central instances make their behavior unpredictable and, therefore, it is difficult to achieve a high level of Quality-of-Service for P2P traffic. Thus, peers are themselves responsible for dealing with these issues by applying so-called self-organization mechanisms to deal with their heterogeneity, unpredictable behavior, and asymmetric resources. This paper discusses and classifies relevant self-organizing aspects of P2P systems, including metrics and mechanisms. Hereby, the key focus is in better understanding on how such self-organizing mechanisms - originally designed to improve the performance of P2P overlays - affect the underlying Internet infrastructure.

Peer-to-peer (P2P) applications have recently attracted a large number of Internet users. Traditional P2P systems however, suffer from inefficiency due to lack of information from the underlay, i.e. the physical network. Although there is a plethora of research on underlay awareness, this aspect of P2P systems is still not clearly structured. In this paper, we provide a taxonomic survey that outlines the different steps for achieving underlay awareness. The main contribution of this paper is presenting a clear picture of what underlay awareness is and how it can be used to build next generation P2P systems. Impacts of underlay awareness and open research issues are also discussed.