Search

The proliferation of the Internet of Things (IoT) and cloud services has given rise to the edge computing paradigm, where data is processed partly or entirely at the edge of the network, rather than solely in the cloud. Edge computing can address problems such as latency, limited battery life of mobile devices, bandwidth costs, security, and privacy [1, 2, 3]. Typical applicable scenarios based on edge computing include video analytics, smart home, smart city, and collaborative edge.With the development of deep learning techniques, research on employing deep learning to develop intelligent edge systems is emerging. In this dissertation, we aim to investigate how deep learning can process data on source-constrained individual edge devices in real time and how deep learning can process data by utilizing collaborative edge devices to provide better services.We build several critical systems, including video analytics, driving anomaly detection, arm posture tracking, and device orientation tracking. In the video analytics system, we combine deep learning with traditional image processing techniques to achieve real-time object detection on mobile devices without offloading. In the driving anomaly detection system, we train deep learning models for driving anomaly detection by leveraging the information from collaborative peer devices to provide better accuracy. In the arm posture tracking system, we employ multitask learning to track the orientation and location of the wrist simultaneously, which significantly improves the latency compared to the conventional methods. In the device orientation tracking system, we develop a deep reinforcement learning framework to train an agent that adjusts the parameters of a conventional orientation tracking method in response to changing environments.As IoT systems continue to grow in complexity and size, preserving training data has become an increasingly important challenge. In our future work, we plan to investigate the use of representation learning to address this issue. By leveraging representation learning techniques, we aim to develop a more robust and efficient method for saving and utilizing training data in IoT systems. This could enable better performance of IoT systems, which in turn could lead to significant improvements in a variety of fields, such as healthcare, transportation, and manufacturing.

The rapidly growing number of edge devices continuously generating data with real-time response constraints coupled with the bandwidth, latency, and reliability issues of centralized cloud computing have made computing near the edge indispensable. As a result, using Field Programmable Gate Arrays (FPGAs) at the edge, due to their unique capabilities that meet the requirements of both high-performance applications and the Internet of Things (IoT) domain, is becoming prevalent. However, designs deployed on these devices suffer from efficiency gap versus custom implementations mainly due to the overhead associated with the FPGAs reconfigurability. This problem is more pronounced in the edge domain, where most devices are battery-powered. In the first part of this dissertation, we identify and overcome the challenges xvi behind the power reduction of FPGA-based applications and propose techniques to lower their energy consumption. Our approach exploits the pessimistic timing margin of the designs to tune the voltage and improves the energy consumption by 66%. An increasing number of edge applications rely on machine learning (ML) algorithms to generate useful insights from data. While modern machine learning techniques – in particular deep neural networks (DNNs) – can produce state-of-the-art results, they often entail substantial memory and compute requirements that may exceed the power and resources available on lightweight error-prone edge devices. Hyperdimensional Computing (HDC) is an emerging lightweight and robust learning paradigm suited for the edge domain that copes with the memory and compute overhead of conventional ML algorithms. The next part of the dissertation proposes efficient FPGA-based and custom hardware implementations of HDC to enable intelligence on devices with limited resources, strict energy constraints, and in noisy environments. The proposed HDC algorithms and accelerators reduce the energy consumption by more than three orders of magnitude compared to other ML solutions, with a comparable or better accuracy. The last part of the dissertation seeks to resolve the privacy concerns of HDC that stem from its reversible algorithm and pose challenges for HDC-based learning and inference. We propose hardware- and communication-efficient techniques that improve the ‘inference’ privacy of HDC by reducing the information of the transferred data while consuming less energy than the non-private baseline. We then show that HDC ‘learning’ can meet tight privacy budgets with negligible accuracy degradation. We also propose a hybrid CNN and HDC model for differentially-private training over image data, which achieves comparable or better accuracy than the state-of-the-art CNN-only methods with more than three orders of magnitude faster training.

The Internet of Things (IoT) integrates heterogeneous devices, ranging from sensors to smartphones, tablets and edge servers, and can provide a variety of services, beyond the traditional Internet. Unfortunately, due to its unprecedented scale and ubiquity, IoT faces a maintainability challenge and a set of interrelated problems. With the emergence of edge computing, IoT devices execute various tasks that consume a significant amount of power to deliver high quality of service, which can drain their battery in short time. High peak power increases the device temperature stress, which worsens the impact of transistors and interconnects reliability degradation mechanisms. Such mechanisms lead to early device failures and are costly to fix. In this dissertation, we focus on novel solutions for energy-efficient and reliability-driven management of IoT systems. We introduce a simulation framework called RelIoT to enable reliability evaluation and analysis in IoT networks, which paves the way for the development of new network management solutions. We develop a dynamic reliability management technique based on computation offloading for IoT edge computing architectures. Our approach achieves 20.5% longer mean time to failure than the next best network management solution. We also present an adaptive and distributed reliability-aware routing protocol using reinforcement learning. We show that our routing protocol can improve reliability of a network up to 73.2% compared to state-of-the-art routing approaches. The main focus inall our solutions is to use device batteries efficiently, satisfy QoS requirements, and improve overall network lifetime by mitigating reliability degradation. Lastly, we complement this to specifically study battery health and associated degradation mechanisms, as the traditional techniques developed for optimizing the energy consumption of networks do not yield optimal battery life. An improvement in network lifetime up to 68.5% can be achieved with our approach compared to energy consumption optimization approaches.

In this paper, a long-term workload management problem for multi-server edge computing with server collaboration is studied. In the considered model, mobile users computation-intensive tasks are generated dynamically over the time and offloaded to associated edge servers according to pre-determined subscription agreements. Upon receiving the subscribed workload, each edge server can then decide to whether participate in server collaboration for enabling workload re-allocation (i.e., workload exchange) with other heterogeneously configured edge servers. Unlike most of the existing work, this paper takes into account both competitions and collaborations among strategic edge servers in sharing their computing capacities. To achieve the equilibrium for each edge server in minimizing its expected cost (including energy consumption, delay, transmission, configuration and pricing costs), a joint optimization is formulated for determining i) its amount of workload to undertake, ii) compensation price charged from peers, and iii) computing speed to adopt. To efficiently solve this problem, we propose a novel cooperative queueing game approach, which integrates a convex optimization, a core cost sharing scheme and a mapping rule. Theoretical analyses and extensive simulations are conducted to evaluate the performance of the proposed solution, and demonstrate its superiority over counterparts.

Driven by the increasing demand of real-time mobile application processing, Multi-access Edge Computing (MEC) has been envisioned as a promising paradigm for pushing computational resources to network edges. In this paper, we investigate an MEC network enabled by Unmanned Aerial Vehicles (UAV), and consider both the multi-user computation offloading and edge server deployment to minimize the system-wide computation cost under dynamic environment, where users generate tasks according to time-varying probabilities. We decompose the minimization problem by formulating two stochastic games for multi-user computation offloading and edge server deployment respectively, and prove that each formulated stochastic game has at least one Nash Equilibrium (NE). Two learning algorithms are proposed to reach the NEs with polynomial-time computational complexities. We further incorporate these two algorithms into a chess-like asynchronous updating algorithm to solve the system-wide computation cost minimization problem. Finally, performance evaluations based on real-world data are conducted and analyzed, corroborating that the proposed algorithms can achieve efficient computation offloading coupled with proper server deployment under dynamic environment for multiple users and MEC servers.

Mobile Edge Computing (MEC) has emerged as a promising paradigm catering to overwhelming explosions of mobile applications, by offloading the compute-intensive tasks to an MEC network for processing. The surging of deep learning brings new vigor and vitality to shape the prospect of intelligent Internet of Things (IoT), and edge intelligence arises to provision real-time deep neural network (DNN) inference services for users. In this paper, we study a novel delay-aware DNN inference throughput maximization problem by accelerating each DNN inference through jointly exploring DNN partitioning and multi-thread parallelism. Specifically, we consider the problem under both offline and online request arrival settings: a set of DNN inference requests is given in advance, and a sequence of DNN inference requests arrives one by one without the knowledge of future arrivals, respectively. We first show that the defined problems are NP-hard. We then devise a novel constant approximation algorithm for the problem under the offline setting. We also propose an online algorithm with a provable competitive ratio for the problem under the online setting. We finally evaluate the performance of the proposed algorithms through experimental simulations. Experimental results demonstrate that the proposed algorithms are promising.

Fog/Edge computing is a novel computing paradigm supporting resource-constrained Internet of Things (IoT) devices by the placement of their tasks on the edge and/or cloud servers. Recently, several Deep Reinforcement Learning (DRL)-based placement techniques have been proposed in fog/edge computing environments, which are only suitable for centralized setups. The training of well-performed DRL agents requires manifold training data while obtaining training data is costly. Hence, these centralized DRL-based techniques lack generalizability and quick adaptability, thus failing to efficiently tackle application placement problems. Moreover, many IoT applications are modeled as Directed Acyclic Graphs (DAGs) with diverse topologies. Satisfying dependencies of DAG-based IoT applications incur additional constraints and increase the complexity of placement problems. To overcome these challenges, we propose an actor-critic-based distributed application placement technique, working based on the IMPortance weighted Actor-Learner Architectures (IMPALA). IMPALA is known for efficient distributed experience trajectory generation that significantly reduces the exploration costs of agents. Besides, it uses an adaptive off-policy correction method for faster convergence to optimal solutions. Our technique uses recurrent layers to capture temporal behaviors of input data and a replay buffer to improve the sample efficiency. The performance results, obtained from simulation and testbed experiments, demonstrate that our technique significantly improves the execution cost of IoT applications up to 30\% compared to its counterparts., Comment: This Paper is accepted in IEEE Transactions on Mobile Computing (TMC), on 23 October 2021

Mobile Network Operators (MNO) can reduce their Capital and Operational Expenditure (CAPEX) and (OPEX) with the help of tower sharing approach by utilizing the physical infrastructure equipped by a third party tower provider to expand their network coverage. Moreover, Computing Resource Providers (CRP) are also setting up their micro-datacenters at tower stations to provide the Multi-access Edge Computing (MEC) services by cooperating with tower providers. Since both the communication and computing services contribute to the task offloading in MEC, the resource allocation has become a challenging problem. In this paper, we formulate the joint uplink, downlink, and computing resources allocation problem in which the objectives of both MNOs and CRP are to minimize their OPEX. The task offloading is modeled as a network of queues where the end-to-end latency is calculated based on the performance of the queue network. Then, the formulated problem is transformed into a Generalized Nash Equilibrium Problem (GNEP) to capture the conflicting interests in the resource allocation among MNOs and CRP. To solve the formulated GNEP efficiently, two decentralized algorithms are proposed by introducing the penalty parameters to the coupling constraints. In addition, the convergence and performance of the algorithms on different parameters are analyzed.

Multi-access edge computing (MEC) has been recently considered in challenging environments lacking available terrestrial infrastructures by extending the computing resources to the air for further enhancing the computation capability of the new aerial user equipment (AUE). Additionally, wireless power transfer (WPT) is a promising solution to prolong the battery lifetime of energy-constrained wireless devices like AUEs. In this paper, we investigate the integration of laser-beamed WPT in the high-altitude platform (HAP) aided MEC systems for the HAP-connected AUEs. By discretizing the three-dimensional coverage space of the HAP, we present a multi-tier tile grid-based spatial structure to provide aerial locations for laser charging. With this setup, we identify a new privacy vulnerability caused by the openness during the air-to-air transmission of WPT signaling messages in the presence of a terrestrial adversary. A privacy-aware laser-powered aerial MEC framework is developed that addresses this vulnerability and enhances the location privacy of AUEs for laser WPT. Specifically, the interaction between the HAP as a defender and the adversary in their tile grid allocation as charging locations to AUEs is formulated as a Colonel Blotto game, which models the competition of the players for limited resources over multiple battlefields for a finite time horizon. Moreover, we derive the mixed-strategy Nash equilibria of the tile grid allocation game for both symmetric and asymmetric tile grids between the defender and the adversary. Simulations results show that the proposed framework significantly outperforms the design baselines with a given privacy protection level in terms of system-wide expected total utilities.

The edge computing domain is notably diverse in its composition. Capabilities including sensing, actuation, and mobility are often constrained by limitations such as those in energy, storage, and hardware-specific implementations. Leveraging the diversity of cloud-edge systems when building applications pose fundamental challenges that constrain expressivity, portability, and reconfigurability. Given this complexity, system tooling is necessary to aid application development and deployment in order to provide a stable runtime foundation by which environment dynamics can be accounted for. This dissertation explores the notion of incorporating self-awareness into autonomous systems spanning the cloud and edge, such that they might observe the constantly changing state of their applications and resources to implicitly adapt for optimized behavior. Key thrusts include (1) DDFlow, a macroprogramming abstraction that organizes distributed applications into a visual dataflow representation, (2) Portkey, an adaptive key-value store that reconfigures data placement based on access patterns, and (3) EdgeRM, a distributed resource manager that unifies a heterogeneous computing cluster spanning high-performance servers and low-power sensors, which leverages a WebAssembly execution model and system interface extension for unified sensing.Ultimately, this research seeks to lift IoT and embedded devices into the general-purpose computing domain, thereby enabling a carry-over of technical contributions pioneered by the distributed systems community. As systems researchers continue to introduce new paradigms in how to design, deploy, and support applications spanning cloud-edge, our community can help manage the difficulty in harnessing the capabilities of these networks.

The operations of Mobile Autonomous Systems (MAS) rely on real-time data analysis. For instance, autonomous vehicles’ navigation requires the low-latency analysis of high resolution images to detect, and avoid, objects. Unfortunately, many Mobile Autonomous Systems (MASs) have constrained computing and energy resources, and the continuous execution of state-of-the-art algorithms is out of their reach. By offloading the processing load to compute-capable device located at the network edge, the edge computing paradigm can mitigate this issue. However in practical real-world settings, the wireless channel connecting the mobile devices to the edge server often presents erratic capacity patterns due to mobility.As a result, the overall delay perceived by the mobile application may be affected by large variations, which in turn harm control.This thesis explores solutions to the problem described above. To this aim, in addition to new concepts and edge offloading strategies, a complete real-world platform – the HyDRA platform, was developed to support design and evaluation, as well as dataset collection.HyDRA is a fully open source software and hardware platform realizing flexible machine learning-empowered computing for MAS. From a hardware perspective, HyDRA is composed of several Unmanned Aerial Vehicles (UAVs) and ground devices collaboratively performing data analysis to accomplish system-wide goals. From a software perspective, the HyDRA middleware enables real-time control of data and task routing within the system, organized as a distributed set of modules transforming the data captured by the MAS into actuable control.Taking HyDRA as a starting point, this thesis makes the following conceptual contributions:‚the end-to-end delay in remote computing settings for MAS – and specifically autonomous quad-copters – were characterized by means of real-world experiments that produced a comprehensive dataset focused on object detection from images. The study considered both Wi-Fi and Long-Term Evolution connectivity, and several embedded computing platforms. The results demonstrates the instability of application level delay even in line-of-sight settings and relatively slow vehicle motion. A framework for the dynamic control of task offloading in MASs with extreme temporal variations is developed. The frameworks is based on a preliminary experimental analysis, which indicates that there is no dominant feature, including obvious features such as channel quality, and that prediction necessitates an ensemble of weaker features. We first mathematically formulate a Redundant Task Offloading Problem. Then, we create predictors that can help manage the resource usage/performance trade-off. Specifically, we propose a myopic predictor as baseline and a DRL-based approach, which operates on a set of features from application, network and device-level components. To the best of our knowledge, this is the first framework addressing the problem of redundant task offloading in MAS with a data-driven approach which efficacy is verified in a real-world testbed and with replicable dataset-based experiments.‚A modeling and optimization framework based on Markov Decision Processes (MDP) was developed to analyze the structural properties of dynamic control strategies determining where information is processed in collaborative computing scenarios for MAS. In this section of the thesis, the focus of control is primarily between local and remote analysis. Using recent split Deep Neural Networks deep neural network (DNN) techniques, the framework also controls at a fine-grain how the analysis task (the DNN in this case) is divided between the mobile device and the edge server based on current system parameters.

The convergence of real-time embedded systems, wireless sensor networks, and machine learning, has fueled the rapid development of the Internet of Things (IoT), engendering new computational workloads and generating unprecedented amounts of streaming data. As a result, the computational infrastructure for IoT is facing challenges imposed by scalability, varying availability and performance, and heterogeneity. Highly concurrent event-driven architectures (EDAs) are one potential technological approach to building large-scale IoT systems since they naturally comprise concurrency, scheduling, and service decoupling and isolation. Moreover, serverless computing is an example of an EDA that has emerged as the next-generation event-driven system to address many of these challenges.In addition, new tiered cloud architectures consisting of low-capability IoT devices, computing and storage resources sited ``at the network edge,'' and public cloud resources provide the opportunity to optimize placement of computation and storage tasks to achieve the performance and reliability requirements of IoT applications. The ``edge cloud'' is a service-hosting technology (like a public cloud) but located at the network edge to enable IoT deployments to take advantage of the spatial locality to optimize resource utilization, reduce required wide-area bandwidth, reduce response latency, improve application fault resilience, and improve security. To maximize its benefits, an efficient scheduling system that intelligently places workloads across IoT, edge cloud, and private/public cloud resources is indispensable. In this thesis, we report our research on building a scalable, event-driven, geo-distributed intelligent scheduling system for heterogeneous IoT devices and applications at the network edge.To achieve the goal, we investigate the efficacy of using a serverless computing platform for tuning machine learning applications in parallel. We also research the usage of serverless computing across the edge and private/public cloud deployments for intelligent scheduling. A third investigation focuses on controlling the system temperature of edge cloud resources by dynamic voltage and frequency scaling (DVFS) to prevent overheating in environments hostile to computational infrastructure. Our work contributes to the corpus of computation offloading research for cyber-physical systems (CPS) that pairs workloads and resources among an available pool of heterogeneous IoT devices, edge cloud resources, and private/public cloud resources.

Network applications involve massive heterogeneous data fusion and analysis. Artificial intelligence can significantly improve the convenience and user experience, but it requires a lot of storage, bandwidth, and computing resources. Multi-access edge computing (MEC) extends intelligence services to IoT devices through offloading approaches and joint processing, which solves the resource bottleneck. However, designing advanced collaboration technology to offload tasks to MEC servers is still challenging. Heuristic algorithms and deep reinforcement learning (DRL) based approaches have been proposed to offload tasks and minimize application latency. However, heuristic algorithms heavily depend on accurate mathematical models for the MEC system, and DRL does not make fair use of the relationship between devices in the MEC graph. To solve this, we propose a task offloading mechanism based on graph neural network (GNN), which can directly learn on graph data with messages passing and aggregation. We propose a Graph Reinforcement Learning-based Offloading (GRLO) framework, which models MEC as an acyclic graph and the offloading policy by graph state migration. GRLO combines GNN with the actor-critic network and trains offloading decision-makers without labels. To efficiently train the GRLO, we propose a method that quickly explores action space and approaches the optimal solution. The numerical results show that the GRLO has lower latency compared to baselines while having generalization ability to new environments and topologies. Moreover, we verified the effectiveness of GRLO on a prototype.

The development of Industrial Internet of Things (IIoT) and Industry 4.0 has completely changed the traditional manufacturing industry. Intelligent IIoT technology usually involves a large number of intensive computing tasks. Resource-constrained IIoT devices often cannot meet the real-time requirements of these tasks. As a promising paradigm, Mobile Edge Computing (MEC) system migrates the computation intensive tasks from resource-constrained IIoT devices to nearby MEC servers, thereby obtaining lower delay and energy consumption. However, considering the varying channel conditions as well as the distinct delay requirements for various computing tasks, it is challenging to coordinate the computing task offloading among multiple users. In this paper, we propose an autonomous partial offloading system for delay sensitive computation tasks in multi-user IIoT MEC systems. Our goal is to provide offloading services with minimum delay for better Quality of Service (QoS). Enlighten by the recent advancement of Reinforcement Learning (RL), we propose two RL based offloading strategies to automatically optimize the delay performance. Specifically, we first implement Q-learning algorithm to provide a discrete partial offloading decision. Then, to further optimize the system performance with more flexible task offloading, the offloading decisions are given as continuous based on Deep Deterministic Policy Gradient (DDPG). The simulation results show that, the Q-learning scheme reduces the delay by 23%, and the DDPG scheme reduces the delay by 30%.

Edge computing can provide high bandwidth and low-latency service for big data tasks by leveraging the edge side’s computing, storage, and network resources. With the development of microservice and docker technology, service providers can flexibly and dynamically cache microservice at the edge side to respond efficiently with limited resources. Automatically caching needed services on the nearest edge nodes and dynamically scheduling users’ requests can realize that computing power and software services flow with the users to provide continuous services. However, achieving the goal needs to overcome many challenges, such as the significant fluctuation of user devices’ requests at the edge side and the lack of collaboration among edge nodes. In this paper, dynamic computing power scheduling and collaborative task scheduling among edge nodes are comprehensively developed. The problem is considered a multi-objective optimization problem, including sequentially minimizing the deadline missing rate of requests and the average task completion time. We propose an adaptive mechanism for dynamically collaborative computing power and task scheduling (ADCS) in the edge environment to solve this problem. It adopts the greedy decision method to schedule computing tasks to meet their deadline requirements. At the same time, it uses the best-fit method to adjust the computing resources according to the changes of users’ requests. The simulation results show that ADCS can decrease the deadline missing rate and reduce the average completion time. Compared with DSR and CoDSR, the deadline missing rate is reduced by 59.91% and 19.95%, respectively. The average completion time is decreased by 37.87%, 6.71%.

With the rapid advance of Internet of Things (IoTs), it is difficult for cloud-centric computing to meet the requirements of low latency and ease of use. As an open and distributed system, edge computing integrates computing, networking, storage, and applications. It provides intelligent services on the edge of an IoT. The edge network is comprised of various wireless and wired networks, and the computing and storage resources of edge nodes are limited. These conditions make the edge network expose to a variety of cyber attacks. Additionally, it is difficult for an IoT edge node to support large-scale network data collection and detection for IoT security. Although big data-enabled intrusion detection algorithms can ensure the high accuracy of intrusion detection systems, it is stressful for resource-limited edge nodes to implement those algorithms in IoTs. Motivated by these challenges, we propose an intelligent intrusion detection algorithm implemented by big data mining based on fuzzy rough set, Generative Adversarial Network (GAN), and Convolutional Neural Network (CNN). In our method, we first propose a fuzzy rough set-based algorithm to perform feature selection for big data via IoTs. Then, we take advantage of the efficient feature extraction capabilities of CNN for implementing intrusion detection based on selected features. Furthermore, after combining CNN and GAN, we propose an intelligent algorithm to realize intrusion detection in a variety of scenarios. Finally, the proposed method is compared with existing methods for evaluation. Simulation results show that our method has up to 4% higher accuracy than existing methods.

Edge computing (EC) offers cloud-like services at the edge of mobile networks to satisfy the delay-sensitive and rapid computation applications in meeting the demands of rapidly increasing mobile devices and other IoTs. EC is known to be constrained with limited resources that its efficacy greatly depends on an effective and efficient resource allocation to provide optimal resources utilization. Focusing on the fact, this paper presents an adaptive resource allocation mechanism, abbreviated as A-PBRA, for effective resources utilization in the EC paradigm. To realize optimal utilization, the available resources are allocated dynamically (adaptability) by considering the nature of the incoming requests. The proposed scheme shall adapt to the resource demands and priorities of the incoming requests. After identifying the received request which can be either priority-based or normal request, each of them is processed with three possibilities. The available resources are thus allocated as per the priorities of the incoming requests to satisfy the constraints accordingly. The proposed mechanism is adaptable to a maximum number of incoming requests along with optimizing the utilization of limited resources at the edge node. Extensive simulations were performed through ifogsim to evaluate the performance of the proposed method. Critical comparisons were made against closely related algorithms and techniques i.e., the NBIHA and the CORA-GT. The simulation results from the proposed scheme optimistically showing that it performed better in terms of resources utilization, average response time, task execution time, and energy consumption.

Edge computing has been recently introduced as a way to bring computational capabilities closer to end users of modern network-based services, in order to support existent and future delay-sensitive applications by effectively addressing the high propagation delay issue that affects cloud computing. However, the problem of efficiently and fairly manage the system resources presents particular challenges due to the limited capacity of both edge nodes and wireless access networks, as well as the heterogeneity of resources and services' requirements. To this end, we propose a techno-economic market where service providers act as buyers, securing both radio and computing resources for the execution of their associated end users' jobs, while being constrained by a budget limit. We design an allocation mechanism that employs convex programming in order to find the unique market equilibrium point that maximizes fairness, while making sure that all buyers receive their preferred resource bundle. Additionally, we derive theoretical properties that confirm how the market equilibrium approach strikes a balance between fairness and efficiency. We also propose alternative allocation mechanisms and give a comparison with the market-based mechanism. Finally, we conduct simulations in order to numerically analyze and compare the performance of the mechanisms and confirm the theoretical properties of the market model., Comment: Corrected typos and figure orientation. in IEEE Transactions on Mobile Computing

The quantitative performance analysis plays a critical role in assessing the capability of Vehicular Edge Computing (VEC) systems to meet the requirements of vehicular applications. However, developing accurate analytical models for VEC systems is extremely challenging due to the unique features of intelligent vehicular applications. Specifically, recent work revealed that the tasks generated by intelligent vehicular applications exhibit a high degree of burstiness, rendering the existing models that were designed based on the assumption of the non-bursty Poisson process unsuitable for VEC systems. To fill this gap, we developed an original analytical model to investigate the performance of VEC systems with bursty task arrivals. To facilitate vehicle cooperation, a new priority-based resource allocation scheme is exploited to schedule the tasks of vehicular applications, which are modelled by a Markov Modulated Poisson Process (MMPP). Next, a multi-state Markov chain is established to investigate the impact of load sharing strategy on the performance of VEC systems. Then, the end-to-end transmission latency is derived based on the proposed model. Comprehensive experiments are conducted to validate the accuracy of this analytical model under various system configurations. Furthermore, the developed model is used as a cost-effective tool to investigate the performance bottleneck of VEC systems.

In Vehicular Edge Computing (VEC) systems, the computing resources of connected Electric Vehicles~(EV) are used to fulfill the low-latency computation requirements of vehicles. However, local execution of heavy workloads may drain a considerable amount of energy in EVs. One promising way to improve the energy efficiency is to share and coordinate computing resources among connected EVs. However, the uncertainties in the future location of vehicles make it hard to decide which vehicles participate in resource sharing and how long they share their resources so that all participants benefit from resource sharing. In this paper, we propose VECMAN, a framework for energy-aware resource management in VEC systems composed of two algorithms: (i) a resource selector algorithm that determines the participating vehicles and the duration of resource sharing period; and (ii) an energy manager algorithm that manages computing resources of the participating vehicles with the aim of minimizing the computational energy consumption. We evaluate the proposed algorithms and show that they considerably reduce the vehicles computational energy consumption compared to the state-of-the-art baselines. Specifically, our algorithms achieve between 7\% and 18\% energy savings compared to a baseline that executes workload locally and an average of 13\% energy savings compared to a baseline that offloads vehicles workloads to RSUs.

We study the problem of Resilient Service Provisioning for Edge computing (RSPE), i.e., how to determine a service placement strategy to maximize the expected overall utility by service provisioning, in the presence of uncertain service failures. RSPE is extremely challenging to tackle, because the explicit expression of its objective function is difficult to obtain, and it is a resilient max-min problem subject to knapsack constraints, which is unexplored so far and cannot be addressed by existing resilient optimization techniques. We first explore the potential properties of the implicit objective function, and reveal that it is monotone submodular under certain conditions. We further prove that the knapsack constraints form a q-independence system constraint, where q>0 is a constant related to the constraints. We propose two novel solutions for the general RSPE and homogeneous case, respectively. Firstly, for the general problem, we propose a “two-step greedy” algorithm achieving constant approximation ratio within polynomial time. Secondly, for the homogeneous case where one of the knapsack constraints reduces to a matroid constraint, we propose an improved “first-greedy-then-local search” polynomial time algorithm achieving better approximation ratio than the previous one. Both extensive simulations and field experiments validate the effectiveness of our proposed algorithms.

Federated learning (FL) has been widely adopted to train machine learning models over massive data in edge computing. However, machine learning faces critical challenges, \eg, data imbalance, edge dynamics, and resource constraints, in edge computing.The existing FL solutions cannot well cope with data imbalance or edge dynamics, and may cause high resource cost. In this paper, we propose an adaptive asynchronous federated learning (AAFL) mechanism. To deal with edge dynamics, the parameter server will aggregate local updated models only from a certain fraction $\alpha$ of all edge nodes in each epoch. Moreover, the system can intelligently vary the number of local updated models for global model aggregation in different epochs with network situations. We then propose experience-driven algorithms based on deep reinforcement learning (DRL) to adaptively determine the optimal value of $\alpha$ in each epoch for two cases of AAFL, single learning task and multiple learning tasks, so as to achieve less completion time of training under resource constraints. Extensive experiments on the classical models and datasets show high effectiveness of the proposed algorithms. Specifically, AAFL can reduce the completion time by about 55\% and improve the learning accuracy by 18\% under resource constraints, compared with the state-of-the-art solutions.

This article dedicates to automatically explore efficient portrait parsing models that are easily deployed in edge computing or terminal devices. In the interest of the tradeoff between the resource cost and performance, we design the multiobjective reinforcement learning (RL)-based neural architecture search (NAS) scheme, which comprehensively balances the accuracy, parameters, FLOPs, and inference latency. Finally, under varying hyperparameter configurations, the search procedure emits a bunch of excellent objective-oriented architectures. The combination of two-stage training with precomputing and memory-resident feature maps effectively reduces the time consumption of the RL-based NAS method, so that we complete approximately 1000 search iterations in two GPU days. To accelerate the convergence of the lightweight candidate architecture, we incorporate knowledge distillation into the training of the search process. This also provides a reasonable evaluation signal to the RL controller that enables it to converge well. In the end, we conduct full training with outstanding Pareto-optimal architectures, so that a series of excellent portrait parsing models (with only approximately 0.3M parameters) is received. Furthermore, we directly transfer the architectures searched on CelebAMask-HQ (Portrait Parsing) to other portrait and face segmentation tasks. Finally, we achieve the state-of-the-art performance of 96.5% MIOU on EG1800 (portrait segmentation) and 91.6% overall F1-score on HELEN (face labeling). That is, our models significantly surpass the artificial network on the accuracy, but with lower resource consumption and higher real-time performance.

The 5G network presents a new horizon with tremendous opportunities for future generation wireless networks. Mobile edge computing (MEC), non-orthogonal multiple access (NOMA), and network slicing (NS) are some of the key enablers for 5G. MEC reduces the latency to a great extent for a wireless network, while NOMA gives access to more users with resource constraints. NS provides users with a better quality of service and network operators with more flexibility. In this work, we propose an NS technique enabled with NOMA for a MEC network. The proposed NS technique improves service latency for MEC users and reduces the unnecessary allocation of radio resources in NOMA. The saved resources can be leveraged to accommodate more users, thus increasing the spectral efficiency of the network. We consider different types of services based on the task completion time of users in this work. The primary focus is to optimize the total energy consumption for wireless uplink transmission for the NOMA-enabled sliced MEC network. We also propose a heuristic algorithm as an alternative to reduce the time and computational complexities of the optimization algorithm and simulate the results extensively to show the effectiveness of our proposed algorithm.

The proliferation of Internet of Things (IoT) data and innovative mobile services has promoted an increasing need for low-latency access to resources such as data and computing services. Mobile edge computing has become an effective computing paradigm to meet the requirement for low-latency access by placing resources and dispatching tasks at the edge clouds near mobile users. The key challenge of such solution is how to efficiently place resources and dispatch tasks in the edge clouds to meet the QoS of mobile users or maximize the platform's utility. In this paper, we study the joint optimization problem of resource placement and task dispatching in mobile edge clouds across multiple timescales under the dynamic status of edge servers. We first propose a two-stage iterative algorithm to solve the joint optimization problem in different timescales, which can handle the varieties among the dynamic of edge resources and/or tasks. We then propose a reinforcement learning (RL) based algorithm which leverages the learning capability of Deep Deterministic Policy Gradient (DDPG) technique to tackle the network variation and dynamic as well. The results from our trace-driven simulations demonstrate that both proposed approaches can effectively place resources and dispatching tasks across two timescales to maximize the total utility of all scheduled tasks.

With the development of MEC services, MEC brokers will emerge to facilitate the purchase and management of resources for individual MEC users. Both data communication and computing resources offered by MEC service providers can be purchased by pay-as-you-go (PAYG) or reserved plans. Besides data and computing plans for each type of resource, we also consider combo plans specifically designed for MEC services covering both resources. In this paper, we propose a smart online aggregated reservation (SOAR) framework for MEC brokers to minimize their cost of reserving resources for multiple users without the knowledge of future demands. In our framework, a task aggregation algorithm is designed to aggregate the users' demands in each PAYG billing cycle to improve the plan utilization, and plan reservation algorithms are proposed to decide when to reserve which plans. The performance gap (competitive ratio) between SOAR and optimal solution which knows all future demands in advance, is analyzed and derived in closed-form. The performance gap is proved to be the minimum among all deterministic online algorithms. Trace-driven simulations verify the cost advantage of our SOAR framework, which can save nearly 40% of cost for users through the brokerage service.

The new generation of edge computing supported industrial cyber-physical system (ICPS) promotes the deep integration of sensing and control. The unknown model is one of the key challenges to characterize their interactions. In most existing works, many efforts have been devoted to overcoming the challenge for the single aspect of sensing and control. However, the industrial revolution puts forward the higher requirements of the overall production performance. To solve this problem, we propose a novel framework for learning-based edge sensing and control co-design. Specifically, the model learning error is first analyzed to bound the actual control performance. Then, the bound is further linked to the sensing design through the bridge of relaxed assumptions of the nonzero initial state and unknown order. Besides, the

Edge computing provides computing capability at close-user proximity to reduce service latency for end users. To improve the efficiency of edge computing infrastructures, geographically-distributed edge datacenters can co-work with each other and with cloud datacenters, forming a new paradigm referred to as cooperative edge-cloud computing. In this context, applications typically run on a virtual machine (VM) that can be replicated at multiple sites, and thus user traffic can be served at all the sites where corresponding VMs reside. For the performance of many applications, latency is a critical parameter. In this work, taking applications latencies as the primary constraint, we model the problem of VM placement and workload assignment as a mixed integer linear program and develop heuristic algorithms accordingly. The goal is to minimize the consumption of information technology (IT) infrastructures for placing VMs in cooperative edge-cloud computing, while meeting the heterogeneous latency demands of different applications. Some preliminary results indicate that edge datacenters resource efficiency can be optimized by proper cross-site VM placement and workload re-direction.

International audience; Internet of Things (IoT) applications often require the processing of data streams generated by devices dispersed over a large geographical area. Traditionally, these data streams are forwarded to a distant cloud for processing, thus resulting in high application end-to-end latency. Recent work explores the combination of resources located in clouds and at the edges of the Internet, called cloud-edge infrastructure, for deploying Data Stream Processing (DSP) applications. Most previous work, however, fails to scale to very large IoT settings. This paper introduces deployment strategies for the placement of DSP applications on to cloud-edge infrastructure. The strategies split an application graph into regions and consider regions with stringent time requirements for edge placement. The proposed Aggregate End-to-End Latency Strategy with Region Patterns and Latency Awareness (AELS+RP+LA) decreases the number of evaluated resources when computing an operator’s placement by considering the communication overhead across computing resources. Simulation results show that, unlike the state-of-the-art, AELS+RP+LA scales to environments with more than 100k resources with negligible impact on the application end-to-end latency.

Mobile edge computing achieves better application responsiveness by offloading tasks from end devices to edge servers installed at the vicinity. Practical scenarios such as post-disaster rescuing and battlefield monitoring make it attractive to use end devices themselves as edge servers. This however introduces a new challenge: Due to mobility and power limitation, the set of edge servers becomes dynamic. As some servers fail, the tasks that run on them will also fail. This paper introduces a new dynamic edge computing model and conducts the first study on robust task offloading which is tolerant to h server failures. We propose online primal-dual algorithms that offload tasks as they arrive. We evaluate the performance of our robust task offloading solutions through extensive simulations based on real task sets. The results show that our proposed solutions can well handle edge dynamics and achieve near optimal throughput (above 95%) compared to the optimal offline benchmark algorithm.

The emerging Federated Learning (FL) enables IoT devices to collaboratively learn a shared model based on their local datasets. However, due to end devices heterogeneity, it will magnify the inherent synchronization barrier issue of FL and result in non-negligible waiting time when local models are trained with the identical batch size. Moreover, the useless waiting time will further lead to a great strain on devices limited battery life. Herein, we aim to alleviate the negative impact of synchronization barrier through adaptive batch size during model training. When using different batch sizes, stability and convergence of the global model should be enforced by assigning appropriate learning rates on different devices. Therefore, we first study the relationship between batch size and learning rate, and formulate a scaling rule to guide the setting of learning rate in terms of batch size. Then we theoretically analyze the convergence rate of global model and obtain a convergence upper bound. On these bases, we propose an efficient algorithm that adaptively adjusts batch size with scaled learning rate for heterogeneous devices to reduce the waiting time and save battery life. We conduct extensive simulations and testbed experiments, and the experimental results demonstrate the effectiveness of our method.

Mobile edge computing (MEC), extending computing services from cloud to edge, is recognized as one of key pillars to facilitate real-time services and tackle backhaul bottleneck. However, it is not economically efficient to attach intensive security appliances to every MEC node to defend application-level DDoS attacks and ensure the availability of services. Thus, we explore the elasticity of security defense among MEC nodes by proposing a COoperative DEfense (CODE) framework for MEC, referred to as CODE4MEC. CODE4MEC aims to adapt to traffic changes by coordinating container-carried defensive resources among cooperative MEC nodes in an automatic way. Towards this aim, we propose four control plane functions to enable a life-cycle management for CODE4MEC, namely, CODE triggering, scheduling, coordination and releasing. However, an effective CODE4MEC requires non-trivial algorithmic schemes, in particular for CODE scheduling and coordination functions. We thus design an online combinatorial auction mechanism for real-time CODE scheduling, and prove a tighter performance bound relative to prior arts. As for CODE coordination, a flow-based traffic and context information coordination scheme is proposed to enable classical defense schemes to work properly and efficiently. Finally, using a combination of real testbed and simulation evaluations, we validate the effectiveness of CODE4MEC.

Due to the excessive concentration of computing resources in the traditional centralized cloud service system, there will be three prominent problems of management confusion, construction cost and network delay. Therefore, we propose to virtualize regional edge computing resources in intelligent buildings as edge service pooling, then presents a hierarchical cloud platform with dual-service pooling structure and a dynamic strategy for the proposed model. The analytic hierarchy process (AHP) based quality of service (QoS) evaluation mechanism and the dynamic normal distribution selection method are adopted for service deployment. And the dynamic inertia particle swarm optimization (DI-PSO) algorithm is employed to realize task scheduling. Furthermore, the cloud platform and existing terminal server group are used to conduct platform structure comparison experiments, and the popular task scheduling algorithms are selected for simulation experiments. Experimental results of platform measurement show that the average service response time of different services can be improved by about 17.3% to 37.4%. The average occupancy ratio of computing resources can be reduced by about 5%. The simulation results show that the earliest completion time of single task list can be decreased by 11.3% to 20.9%, and the makespan of 100 task lists can be improved by 0.3 times.

Mobile edge computing is a new paradigm in which the computing and storage resources are placed at the edge of the Internet. Data placement and retrieval are fundamental services of mobile edge computing when a network of edge clouds collaboratively provide data services. However existing methods such as distributed hash tables (DHTs) are not enough to achieve efficient data placement and retrieval services for cooperative edge clouds. This paper presents GRED, a novel data placement and retrieval service for mobile edge computing, which is efficient in not only the load balance but also routing path lengths and forwarding table sizes. GRED utilizes the programmable switches to support a virtual-space based DHT with only one overlay hop. Data location can be easily implemented on top of the GRED. We implement GRED in a P4 prototype, which provides a simple and efficient solution. Results from theoretical analysis, simulations, and experiments show that GRED can efficiently balance the load of edge clouds, and can fast answer data queries due to its low routing stretch.

Innovative edge devices (e.g., smartphones, IoT devices) are becoming much more pervasive in our daily lives. With powerful sensing and computing capabilities, users can generate massive amounts of data. A new business model has emerged where data producers can sell their data to consumers directly to make money. However, how to protect the profit of the data producer from rogue consumers that may resell without authorization remains challenging. In this paper, we propose a smart-contract based protocol to protect the profit of the data producer while allowing consumers to resell the data legitimately. The protocol ensures the revenue is shared with the data producer over authorized reselling, and detects any unauthorized reselling. We also introduce a data relay process that can enhance data accessibility in wireless edge networks. We formulate a revenue sharing problem to maximize the profit of both the data producer and resellers/relayers. We formulate the problem into a two-stage Stackelberg game and determine a ratio to share the reselling revenue between the data producer and resellers/relayers. Extensive simulations show that with resellers and relayers, our mechanism can achieve up to 49.5% higher profit for the data producer and resellers/relayers.

In this paper, a novel data offloading decision-making framework is proposed, where users have the option to partially offload their data to a complex Multi-access Edge Computing (MEC) environment, consisting of both ground and UAV-mounted MEC servers. The problem is treated under the perspective of risk-aware user behavior as captured via prospect-theoretic utility functions, while accounting for the inherent computing environment uncertainties. The UAV-mounted MEC servers act as a common pool of resources with potentially superior but uncertain payoff for the users, while the local computation and ground server alternatives constitute safe and guaranteed options, respectively. The optimal user task offloading to the available computing choices is formulated as a maximization problem of each user's satisfaction, and confronted as a non-cooperative game. The existence and uniqueness of a Pure Nash Equilibrium (PNE) are proven, and convergence to the PNE is shown. Detailed numerical results highlight the convergence of the system to the PNE in few only iterations, while the impact of user behavior heterogeneity is evaluated. The introduced framework's consideration of the user risk-aware characteristics and computing uncertainties, results to a sophisticated exploitation of the system resources, which in turn leads to superior users' experienced performance compared to alternative approaches.

In recent years, edge computing has become an increasingly popular computing paradigm to enable real-time data processing and mobile intelligence. Edge computing allows computing at the edge of the network, where data is generated and distributed at the nearby edge servers to reduce the data access latency and improve data processing efficiency. One of the key challenges in data-intensive edge computing is how to place the data at the edge clouds effectively such that the access latency to the data is minimized. In this paper, we study such a data placement problem in edge computing while different data items have diverse popularity. We propose a popularity based placement method which maps both data items and edge servers to a virtual plane and places or retrieves data based on its virtual coordinate in the plane. We then further propose additional placement strategies to handle load balancing among edge servers via either offloading or data duplication. Simulation results show that our proposed strategies efficiently reduce the average path length of data access and the load-balancing strategies indeed provide an effective relief of storage pressures at certain overloaded servers.

Generally the accelerators type identifies the edge computing performance hugely and it includes Integrated Circuit (IC), Graphical Processing Unit (GPU) and Field Programmable Gate Array (FPGA). Among these accelerator FPGAs are widely used in several edge computing areas since it has the ability to reconfigure and has high energy efficient. The consequence also comes out with the paradigm of scheduling efficiency on CPU’s (software tasks) and FPGA (hardware tasks). However the conventional schemes not entirely exploited the variations among hardware and software tasks which results in low scheduling efficiency. Therefore a task scheduling system is proposed on basis of fully active reconfigurable system. The scheduling efficiency for the proposed scheme can be increased by switching the task overheads and by predicting the execution time for hardware with minimum task switching times.

It is widely acknowledged that the success of deep learning is built on large-scale training data and tremendous computing power. However, the data and computing power are not always available for many real-world applications. In this paper, we address the machine learning problem where it lacks training data and limits computing power. Specifically, we investigate domain adaptation which is able to transfer knowledge from one labeled source domain to an unlabeled target domain, so that we do not need much training data from the target domain. At the same time, we consider the situation that the running environment is confined, e.g., in edge computing the end device has very limited running resources. Technically, we present the Faster Domain Adaptation (FDA) protocol and further report two paradigms of FDA: early stopping and amid skipping. The former accelerates domain adaptation by multiple early exit points. The latter speeds up the adaptation by wisely skip several amid neural network blocks. Extensive experiments on standard benchmarks verify that our method is able to achieve the comparable and even better accuracy but employ much less computing resources. To the best of our knowledge, there are very few works which investigated accelerating knowledge adaptation in the community.

We consider a distributed computing framework where the distributed nodes have different communication capabilities, motivated by the heterogeneous networks in data centers and mobile edge computing systems. Following the structure of MapReduce, this framework consists of Map computation phase, Shuffle phase, and Reduce computation phase. The Shuffle phase allows distributed nodes to exchange intermediate values, in the presence of heterogeneous communication bottlenecks for different nodes (heterogeneous communication load constraints). For this setting, we characterize the minimum total computation load and the minimum worst-case computation load in some cases, under the heterogeneous communication load constraints. While the total computation load depends on the sum of the computation loads of all the nodes, the worst-case computation load depends on the computation load of a node with the heaviest job. We show an interesting insight that, for some cases, there is a tradeoff between the minimum total computation load and the minimum worst-case computation load, in the sense that both cannot be achieved at the same time. The achievability schemes are proposed with careful design on the file assignment and data shuffling. Beyond the cut-set bound, a novel converse is proposed using the proof by contradiction. For the general case, we identify two extreme regimes in which the scheme with coding and the scheme without coding are optimal, respectively.

In this paper, we study a novel service reliability augmentation problem in an Edge Computing network, where mobile users request network services with a service function chain (SFC) and reliability expectation requirements. To enhance the service reliability of user requests, it is a common practice to make use of redundant network function (VNF) instance placement in case their primary VNF instances fail. To this end, we first formulate a novel service reliability augmentation problem for each admitted request and show the NP-hardness of the problem. We then provide an admission framework of user requests by placing primary VNF instances of network functions in the SFC to different cloudlets, and then develop a randomized algorithm with a good approximation ratio and high probability for the service reliability augmentation problem, at the expense of moderate resource constraint violations. We also devise a deterministic heuristic for the problem without any resource violation. We thirdly extend the proposed algorithm for a single request to the service reliability augmentation problem for a set of requests. We finally evaluate the performance of the proposed algorithms through experimental simulations. Experimental results demonstrate that the proposed algorithms are promising and superior to their analytical counterparts.