Edge artificial intelligence (AI) is an innovative computing paradigm that aims to shift the training and inference of machine learning models to the edge of the network. This paradigm offers the opportunity to significantly impact our everyday lives with new services such as autonomous driving and ubiquitous personalized health care. Nevertheless, bringing intelligence to the edge involves several major challenges, which include the need to constrain model architecture designs, the secure distribution and execution of the trained models, and the substantial network load required to distribute the models and data collected for training. In this article, we highlight key aspects in the development of edge AI in the past and connect them to current challenges. This article aims to identify research opportunities for edge AI, relevant to bring together the research in the fields of artificial intelligence and edge computing.@en

The Starlink network from SpaceX stands out as the only commercial LEO network with over 2M+ customers and more than 4000 operational satellites. In this paper, we conduct a first-of-its-kind extensive multi-faceted analysis of Starlink performance leveraging several measurement sources. First, based on 19.2M crowdsourced M-Lab speed tests from 34 countries since 2021, we analyze Starlink global performance relative to terrestrial cellular networks. Second, we examine Starlink's ability to support real-time latency and bandwidth-critical applications by analyzing the performance of (i) Zoom conferencing, and (ii) Luna cloud gaming, comparing it to 5G and fiber. Third, we perform measurements from Starlink-enabled RIPE Atlas probes to shed light on the last-mile access and other factors affecting its performance.Finally, we conduct controlled experiments from Starlink dishes in two countries and analyze the impact of globally synchronized "15-second reconfiguration intervals'' of the satellite links that cause substantial latency and throughput variations. Our unique analysis paints the most comprehensive picture of Starlink's global and last-mile performance to date.@en

Researchers have already begun experimenting with next-generation cellular technologies and algorithms to enable use cases that lie beyond the scope of the current 5G standard, e.g. XR, smart factories, AI networks ops, etc. The common denominator requirement of such scenarios is the joint (coupled) operation of radio channel and edge computing resources within the core network. While there are numerous tools that allow experimenting with various aspects of radio resource management and computing resource management individually, there is a lack of solutions that enable researchers to prototype and evaluate applications and technologies dependent on both aspects simultaneously. In this work, we present nextGSIM, a 5G and beyond network simulator that realistically models the radio access network and edge network jointly to provide an end-to-end service to various user devices running microservice-based application workloads. We detail our design decisions and modular architecture of nextGSIM which resembles real-world setup of cellular networks, enabling effective and detailed simulations of resource management algorithms. We demonstrate the effectiveness and capabilities of nextGSIM through indoor factory case study wherein we evaluate widely regarded radio and edge resource management algorithms. We compare these against a joint radio-compute scheduler which emphasizes the need and benefits of joint resource allocation decision making, which is only possible through tools such as nextGSIM.

@en

Not all research leads to fruitful results; trying new ways or methods may surpass state of the art, but sometimes the hypothesis is not proven, the improvement is insignificant, or the system fails because of a design error done years ago in previous works. In a systems discipline like pervasive computing, there are many sources of errors, from hardware issues over communication channels to heterogeneous software environments. However, failure to succeed is not a failure to progress. It is essential to create platforms for sharing insights, experiences, and lessons learned when conducting research in pervasive computing so that the same mistakes are not repeated. And sometimes, a problem is a symptom of discovering new research challenges. Based on the collective input of the First International Workshop on Negative Results in Pervasive Computing (PerFail 2022), co-located with the 20th International Conference on Pervasive Computing and Communications (PerCom 2022), this article presents a comprehensive discussion on perspectives on publishing negative results, useful failures, and lessons learned in pervasive computing.@en

Mobile Augmented Reality (AR) is gaining traction as a compelling application due to recent advancements in hardware and software. Previous studies have suggested that distributing AR services on an edge computing infrastructure can offer significant performance benefits, especially for consolidating concurrent clients. In this study, we shed light on several research challenges directly impacting the effective integration of distributed AR and edge computing. Specifically, we conduct extensive experiments by deploying our distributed stream processing-based AR pipeline, scAtteR, on a representative edge-cloud infrastructure managed by the Oakestra framework. We uncover several unapparent challenges that inhibit the effective marriage of distributed AR when deployed on edge and demonstrate the potential improvements through scAtteR++. We offer valuable insights and best practices to the growing AR research community, specifically those interested in leveraging edge and public cloud technologies for large-scale AR operations.@en

Recent industrial advancements introduce novel safety-critical applications for commercial networks. Remote Piloting (RP) Aerial Vehicles (AVs) is an example application, where reliable wireless connectivity is key to ensure safe operations in the sky. Jointly utilizing cellular and satellite networks can enable robust Multipath (MP) communications; however, their usage must be orchestrated efficiently toward application requirements. In this work, we investigate the MP communications performance of cellular and Low-Earth-Orbit (LEO) satellite links with respect to the Quality-of-Service (QoS) requirements of RP operations. Using MP-Transmission Control Protocol (MPTCP) and MP-Datagram Congestion Control Protocol (MP-DCCP), we evaluate various transport layer configurations to efficiently orchestrate both links and to support the application requirements. For this purpose, we develop an end-to-end MP emulation testbed that can provide means to realistically emulate cellular and LEO links with MPTCP and MP-DCCP. We run bi-direction al RP traffic over our testbed and measure the MP performance using different schedulers and Congestion Control (CC) algorithms. The results show that the flow size largely influences the individual path utilization due to high LEO link-layer losses. Moreover, excessive retransmissions occur on the MPTCP layer due to Head-of-Line (HoL) blocking from asymmetric link conditions. Using MP-DCCP without retransmissions helps avoid late arrivals and can meet the 99.999% communication reliability demand.

@en

Edge computing enables developers to deploy their services on compute resources deployed closer to the users. The abstraction requires powerful orchestration capabilities and the resolution of complex optimization problems. While edge computing is a consistently growing trend, the community (research and industry) still largely embraces adaptations and extensions of existing cloud technologies that have been proven ineffective on edge (e.g. Kubernetes). In this work, we present Oakestra, a novel hierarchical orchestration framework specifically designed for supporting service operation over heterogeneous edge infrastructures. In this demonstration, we showcase the various features and operations of Oakestra using our latency-critical augmented reality (AR) application.@en

Quality of Service (QoS) is a crucial mechanism where the network manages its finite resources to meet the demands promised for some applications at the cost of forsaking some other applications' needs. Information Centric Networking (ICN), although great at features such as name based addressing, decoupling host from data, in network caching, etc., still offers best effort service. This restricts the network suitability for safety critical applications such as tele-operated driving or real-time multimedia applications that often demand guaranteed bandwidth and delay. Reserving sufficient resources as well as establishing an admission control mechanism is one way to ensure performance guarantees. We present a paper emphasizing the need for a "better than best-effort service" in Named Data Networking (NDN), the functionalities to fulfil for a QoS mechanism and the open challenges in establishing resource reservation in NDN.@en

Human mobility shapes our daily lives, our urban environment and even the trajectory of a global pandemic. While various aspects of human mobility and inter-personal contact duration have already been studied separately, little is known about how these two key aspects of our daily lives are fundamentally connected. Better understanding of such interconnected human behaviors is crucial for studying infectious diseases, as well as opportunistic content forwarding. To address these deficiencies, we conducted a study on a mobile social network of human mobility and contact duration, using data from 71 persons based on GPS and Bluetooth logs for 2 months in 2018. We augment these data with location APIs, enabling a finer granular characterization of the users’ mobility in addition to contact patterns. We model stops durations to reveal how time-unbounded-stops (e.g., bars or restaurants) follow a log-normal distribution while time-bounded-stops (e.g., offices, hotels) follow a power-law distribution. Furthermore, our analysis reveals contact duration adheres to a log-normal distribution, which we use to model the duration of contacts as a function of the duration of stays. We further extend our understanding of contact duration during trips by modeling these times as a Weibull distribution whose parameters are a function of trip length. These results could better inform models for information or epidemic spreading, helping guide the future design of network protocols as well as policy decisions.@en

Application domains such as automotive and the Internet of Things may benefit from in-network computing to reduce the distance data travels through the network and the response time. Information Centric Networking (ICN) based compute frameworks such as Named Function Networking (NFN) are promising options due to their location independence and loosely-coupled communication model. However, unlike current operations, such solutions may benefit from orchestration across the compute nodes to use the available resources in the network better. In this paper, we adopt the State Vector Synchronization (SVS), an application dataset synchronization protocol in ICN, to enhance the neighborhood knowledge of in-network compute nodes in a distributed fashion. As such, we design distributed coordination for in-network computation (DICer) that assists the service deployments by improving the resolution of compute requests. We evaluate the performance of DICer against NFN and observe an increase in the resource utilization at the edge and a reduction in the request completion time.@en

Widespread adoption of mobile augmented reality (AR) and virtual reality (VR) applications depends on their smoothness and immersiveness. Modern AR applications applying computationally intensive computer vision algorithms can burden today's mobile devices, and cause high energy consumption and/or poor performance. To tackle this challenge, it is possible to offload part of the computation to nearby devices at the edge. However, this calls for smart task placement strategies in order to efficiently use the resources of the edge infrastructure. In this paper, we introduce Nimbus --- a task placement and offloading solution for a multi-tier, edge-cloud infrastructure where deep learning tasks are extracted from the AR application pipeline and offloaded to nearby GPU-powered edge devices. Our aim is to minimize the latency experienced by end-users and the energy costs on mobile devices. Our multifaceted evaluation, based on benchmarked performance of AR tasks, shows the efficacy of our solution. Overall, Nimbus reduces the task latency by ~4x and the energy consumption by ~77% for real-time object detection in AR applications. We also benchmark three variants of our offloading algorithm, disclosing the trade-off of centralized versus distributed execution.@en

Edge Intelligence (EI) is an emerging computing and communication paradigm that enables Artificial Intelligence (AI) functionality at the network edge. In this article, we highlight EI as an emerging and important field of research, discuss the state of research, analyze research gaps and highlight important research challenges with the objective of serving as a catalyst for research and innovation in this emerging area. We take a multidisciplinary view to reflect on the current research in AI, edge computing, and communication technologies, and we analyze how EI reflects on existing research in these fields. We also introduce representative examples of application areas that benefit from, or even demand the use of EI.@en

Edge computing is an attractive platform where applications, previously hosted in the cloud, shift parts of their workload on resources closer to the users. The field is still in its nascent stages with significant ongoing innovation in small form-factor hardware designed to operate at the edge. However, the increased hardware heterogeneity at the edge makes it difficult for application developers to determine if their workloads will operate as desired. Simultaneously, edge providers have to make expensive deployment choices for the "correct" hardware that will remain suitable for the near future. We present ComB, an application-oriented benchmarking suite for edge that assists early adopters in evaluating the suitability of an edge deployment. ComB is flexible, extensible, and incorporates a microservice-based video analytics pipeline as default workload to measure underlying hardware's compute and networking capabilities accurately. Our evaluation on a heterogeneous testbed shows that ComB enables both providers and developers to understand better the runtime capabilities of different hardware configurations for supporting operations of applications designed for the edge.@en

Based on the collective input of Dagstuhl Seminar (21342), this paper presents a comprehensive discussion on AI methods and capabilities in the context of edge computing, referred as Edge AI. In a nutshell, we envision Edge AI to provide adaptation for data-driven applications, enhance network and radio access, and allow the creation, optimisation, and deployment of distributed AI/ML pipelines with given quality of experience, trust, security and privacy targets. The Edge AI community investigates novel ML methods for the edge computing environment, spanning multiple sub-fields of computer science, engineering and ICT. The goal is to share an envisioned roadmap that can bring together key actors and enablers to further advance the domain of Edge AI. @en

Emerging Remote Piloting (RP) operations of electrified Unmanned Aerial Vehicles (UAVs) demand low-latency and high-quality video delivery to conduct safe operations in the low-altitude airspace. Although cellular networks are one of the prominent candidates to provide connectivity for such operations, their ground-centric nature limits their capabilities in achieving seamless and reliable aerial connectivity. In this paper, we study the feasibility of supporting RP operations with low latency and high-quality video delivery over commercial cellular networks. By setting up an adaptive bitrate video transmission pipeline with the Google Congestion Control (GCC) and Self-Clocked Rate Adaptation for Multimedia (SCReAM) Congestion Control (CC) algorithms, we analyze the video delivery performance for the RP application requirements and compare the performance of GCC and SCReAM against constant bitrate video delivery. Our results show that low-latency video delivery with < 300 ms playback latency between full-HD and 4K resolution can be maintained up to about 95% of the time in the air. While static bitrate video delivery outperforms adaptive streaming in urban location with abundant link capacity, the latter becomes advantageous in rural locations, where the link capacity is affected by fluctuations. Although the study’s findings highlight the capabilities of cellular networks in delivering low-latency video for a safety-critical aerial service, we also discuss the potential improvements and future research challenges for enabling safe operations and meeting the service requirements using cellular networks. We release our collected traces and the video transmission pipeline as open-source to facilitate research in this field.

@en

Edge computing promises to bring computation close to the end-users to support emergent applications such as virtual reality. However, the computational capacity at the edge of the network is currently limited. To become a pervasive paradigm, edge computing needs highly dispersed decentralized deployments, that, contrary to cloud, cannot benefit from economies of scale. In this situation, crowdsourcing appears attractive - there are plenty of computing devices at the disposal of the general public, and these devices are located exactly where computing power is needed the most - at the edge of the network. Crowdsourcing has been a success maker for scientific computing projects, e.g., SETI@home, or distributed ledger systems empowering decentralized finance. However, as of now, there is no crowdsourced system that addresses the needs of edge computing. In this position paper, we aim to identify the causes of this shortcoming, analyze the potential ways to overcome it, and outline future directions.@en

In the early days of cloud computing, datacenters were sparsely deployed at distant locations far from end-users with high end-to-end communication latency. However, today’s cloud datacenters have become more geographically spread, the bandwidth of the networks keeps increasing, pushing the end-users latency down. In this paper, we provide a comprehensive cloud reachability study as we perform extensive global client-to-cloud latency measurements towards 189 datacenters from all major cloud providers. We leverage the well-known measurement platform RIPE Atlas, involving up to 8500 probes deployed in heterogeneous environments, e.g., home and offices. Our goal is to evaluate the suitability of modern cloud environments for various current and predicted applications. We achieve this by comparing our latency measurements against known human perception thresholds and are able to draw inferences on the suitability of current clouds for novel applications, such as augmented reality. Our results indicate that the current cloud coverage can easily support several latency-critical applications, like cloud gaming, for the majority of the world’s population.@en

Cloud computing has seen continuous growth over the last decade. The recent rise in popularity of next-generation applications brings forth the question: "Can current cloud infrastructure support the low latency requirements of such apps?" Specifically, the interplay of wireless last-mile and investments of cloud operators in setting up direct peering agreements with ISPs globally to current cloud reachability and latency has remained largely unexplored.This paper investigates the state of end-user to cloud connectivity over wireless media through extensive measurements over six months. We leverage 115,000 wireless probes on the Speed-checker platform and 195 cloud regions from 9 well-established cloud providers. We evaluate the suitability of current cloud infrastructure to meet the needs of emerging applications and highlight various hindering pressure points. We also compare our results to a previous study over RIPE Atlas. Our key findings are: (i) the most impact on latency comes from the geographical distance to the datacenter; (ii) the choice of a measurement platform can significantly influence the results; (iii) wireless last-mile access contributes significantly to the overall latency, almost surpassing the impact of the geographical distance in many cases. We also observe that cloud providers with their own private network backbone and direct peering agreements with serving ISPs offer noticeable improvements in latency, especially in its consistency over longer distances.@en

Amazon Web Services (AWS) offers transient virtual servers at a discounted price as a way to sell unused spare capacity in its data centers. Although transient servers are very appealing as some instances have up to 90% discount, they are not bound to regular availability guarantees as they are opportunistic resources sold on the spot market. In this paper, we present SPA, a framework that remarkably increases the spot instance reliability over time due to insights gained from the analysis of historical data, such as cross-region price variability and intervals between evictions. We implemented the SPA reliability strategy, evaluated them using over one year of historical pricing data from AWS, and found out that we can increase the transient instance lifetime by adding a pricing overhead of 3.5% in the spot price in the best scenario.

@en

Edge computing has received significant attention from both academic and industrial research circles. The paradigm aims to decentralize the existing cloud infrastructure by incorporating resources co-located alongside its client. Researchers have also proposed solutions for a fully decentralized crowdsourced compute paradigm enabled by Distributed Ledger Technologies (DLTs). This paper investigates the rationale behind DLTs over crowdsourced resource marketplaces to support the requirements of latency-critical applications targeted by edge computing. We develop a fully configurable NEtworked Blockchain emULAtor, or NEBULA, to scrutinize the internal performance bottlenecks of DLTs. We evaluate two blockchain categories - proof-based (popularly used in Bitcoin, Ethereum) and hybrid consensus and find that the enabling factor of DLTs -scale - is also its primary latency contributor. We show that, in reality, the latency overheads due to DLT operation far exceed the operational requirements of edge applications.@en