Edge Resources Research Articles

Learning to Coordinate in Mobile-Edge Computing for Decentralized Task Offloading

Edge servers, which are located in close proximity to mobile users, have become emerging components for computation offloading in multiple Internet of Things (IoT) applications. As the edge resources are limited and shared among multiple mobile users, it is crucial for the users to choose appropriate edge server for task offloading, so that their cumulative utility can be maximized. Reinforcement learning (RL) algorithms, which are sequential and model-free, have been widely considered. However, it is still a critical challenge to coordinate the mobile users in a decentralized way. In this work, we propose a novel framework of Multiagent RL by learning to coordinate. The main idea is to introduce an additional “virtual” agent at the edge, which learns to broadcast public messages to the mobile users at each interval. We then enforce positive correlation between each user’s offloading policy and the message. The underlying intuition is that the message can contain information of edge resources and other users’ policies. Therefore, it is expected that the decentralized users can make coordinated decisions. Theoretical analysis shows that our algorithm can converge to equilibrium points under certain mild assumptions. In the experiments, our approach outperforms other baselines significantly in different scenarios. In addition, the results show that the broadcast message plays a very important role in coordinating the mobile users.

AGENT BASED APPROACH FOR TASK OFFLOADING IN EDGE COMPUTING

Due to limited resource capacity in the edge network and a high volume of tasks offloaded to edge servers, edge resources may be unable to provide the required capacity for serving all tasks. As a result, some tasks should be moved to the cloud, which may cause additional delays. This may lead to dissatisfaction among users of the transferred tasks. In this paper, a new agent-based approach to decision-making is presented about which tasks should be transferred to the cloud and which ones should be served locally. This approach tries to pair tasks with resources such that a paired resource is the most preferred resource by the user or task among all available resources. We demonstrate that reaching a Nash Equilibrium point can satisfy the aforementioned condition. A game-theoretic analysis is included to demonstrate that the presented approach increases the average utility of the user and their level of satisfaction.

Dynamic Data Collection and Neural Architecture Search for Wireless Edge Intelligence Systems

With the booming development of Internet of things (IoT) devices and machine learning (ML) technique, edge machine learning is emerging to process the enormous sampled data for realizing intelligent applications at the network edge. With limited edge resources, a well-structured neural network and numerous training data are the two main factors that affect the performance of edge machine learning. In this paper, we cooperatively optimize the data collection and the neural architecture to minimize the energy consumption of devices and the error on a specific task. We derive the Rademacher complexity bounds theoretically to evaluate the generalization error of the neural architectures in the search space and then formulate the optimization problem accordingly. Then we develop a scheme to solve the problem that dynamically performs the data collection based on policy gradient reinforcement learning and the parameter-sharing neural architecture search (NAS) algorithm. By this way, the transmission power of each device can be adjusted based on the data quality assessed by the NAS result in each round to effectively collect data. And with the growing high-quality data, the NAS algorithm can gradually find the optimal architecture for the task. Experimental results show that the neural architectures found by the proposed algorithm outperform the existing architectures while saving energy in the device.

A Dynamic Multi-Attribute Resource Bidding Mechanism with Privacy Protection in Edge Computing

In edge computing, a reasonable edge resource bidding mechanism can enable edge providers and users to obtain benefits in a relatively fair fashion. To maximize such benefits, this paper proposes a dynamic multi-attribute resource bidding mechanism (DMRBM). Most of the previous work mainly relies on a third-party agent to exchange information to gain optimal benefits. It is worth noting that when edge providers and users trade with third-party agents which are not entirely reliable and trustworthy, their sensitive information is prone to be leaked. Moreover, the privacy protection of edge providers and users must be considered in the dynamic pricing/transaction process, which is also very challenging. Therefore, this paper first adopts a privacy protection algorithm to prevent sensitive information from leakage. On the premise that the sensitive data of both edge providers and users are protected, the prices of providers fluctuate within a certain range. Then, users can choose appropriate edge providers by the price-performance ratio (PPR) standard and the reward of lower price (LPR) standard according to their demands. The two standards can be evolved by two evaluation functions. Furthermore, this paper employs an approximate computing method to get an approximate solution of DMRBM in polynomial time. Specifically, this paper models the bidding process as a non-cooperative game and obtains the approximate optimal solution based on two standards according to the game theory. Through the extensive experiments, this paper demonstrates that the DMRBM satisfies the individual rationality, budget balance, and privacy protection and it can also increase the task offloading rate and the system benefits.

A Full Dive Into Realizing the Edge-Enabled Metaverse: Visions, Enabling Technologies, and Challenges

Dubbed “the successor to the mobile Internet,” the concept of the Metaverse has grown in popularity. While there exist lite versions of the Metaverse today, they are still far from realizing the full vision of an immersive, embodied, and interoperable Metaverse. Without addressing the issues of implementation from the communication and networking, as well as computation perspectives, the Metaverse is difficult to succeed the Internet, especially in terms of its accessibility to billions of users today. In this survey, we focus on the edge-enabled Metaverse to realize its ultimate vision. We first provide readers with a succinct tutorial of the Metaverse, an introduction to the architecture, as well as current developments. To enable ubiquitous, seamless, and embodied access to the Metaverse, we discuss the communication and networking challenges and survey cutting-edge solutions and concepts that leverage next-generation communication systems for users to immerse as and interact with embodied avatars in the Metaverse. Moreover, given the high computation costs required, e.g., to render 3D virtual worlds and run data-hungry artificial intelligence-driven avatars, we discuss the computation challenges and cloud-edge-end computation framework-driven solutions to realize the Metaverse on resource-constrained edge devices. Next, we explore how blockchain technologies can aid in the interoperable development of the Metaverse, not just in terms of empowering the economic circulation of virtual user-generated content but also to manage physical edge resources in a decentralized, transparent, and immutable manner. Finally, we discuss the future research directions towards realizing the true vision of the edge-enabled Metaverse.

In-network Placement of Reusable Computing Tasks in an SDN-based Network Edge

Edge computing is aimed to support compute-intensive data-hungry interactive applications which can hardly run on resource-constrained consumer devices and may suffer from running in the cloud due to the long data transfer delay. The edge network nodes' heterogeneous and limited (compared to the cloud) capabilities make the computing task placement a challenge. In this paper, we propose a novel <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">in-network</i> task placement strategy aimed at minimizing the edge network resources usage. The proposal specifically accounts for <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">time-limited reusable</i> computing tasks, i.e., tasks whose output can be cached to serve requests from different consumers for a certain time. Caching such results, during their time validity, achieves the twofold benefit of reducing the service provisioning time and improving the edge resource utilization, by avoiding redundant computations and data exchange. The devised strategy is implemented as a network application of a Software-defined Networking Controller in charge of overseeing the edge domain. We formulate the optimal task placement through an integer linear programming problem, and we define an efficient heuristic algorithm that well approximates the solution achieved through a standard optimal solver. Achieved results show that the proposal successfully meets the targeted objectives in a wide variety of simulated scenarios, by outperforming benchmark solutions.

Microservices From Cloud to Edge: An Analytical Discussion on Risks, Opportunities and Enablers

Microservice applications are made of many interacting microservices that can leverage different cloud servers to distribute the computational load horizontally, so these applications are considered cloud-native. The emergence of edge computing has prompted the research of placement strategies that consider the possibility of instantiating microservices in cloud and edge data centers to improve specific performance metrics, such as the time the application spent serving a user request, also referred to as user delay or makespan. Microservice applications differ in many architectural properties, such as the number of microservices, the number of them involved per request, their dependency graph, and the presence of centralized databases. We found that these properties influence the possibility of a placement strategy to exploit edge resources to reduce user delay; therefore, they constitute an application-level optimization space that, however, has not yet been explored in the literature. Accordingly, this paper contributes to filling this knowledge gap by answering the following question: which are the architectural properties of a microservice application that enable it to be more edge-native, that is, to better reduce user delay by using edge computing? Our results are based on a new analytical modeling of the average user delay provided by a microservice application deployed on cloud/edge resources and a new heuristic placement strategy that take into account key aspects of the application, and the supporting cloud and network environment, not considered by previous literature works.

Joint Optimization Across Timescales: Resource Placement and Task Dispatching in Edge Clouds

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.

Towards a lightweight task scheduling framework for cloud and edge platform

Mobile devices are becoming ubiquitous in our daily lives, but they have limited computational capacity. Thanks to the advancement in the network infrastructure, task offloading from resource-constrained devices to the near edge and the cloud becomes possible and advantageous. Complete task offloading is now possible to almost limitless computing resources of public cloud platforms. Generally, the edge computing resources support latency-sensitive applications with limited computing resources, while the cloud supports latency-tolerant applications. This paper proposes one lightweight task-scheduling framework from cloud service provider perspective, for applications using both cloud and edge platforms. Here, the challenge is using edge and cloud resources efficiently when necessary. Such decisions have to be made quickly, with a small management overhead. Our framework aims at solving two research questions. They are: (i) How to distribute tasks to the edge resource pools and multi-clouds? (ii) How to manage these resource pools effectively with low overheads? To answer these two questions, we examine the performance of our proposed framework based on Reliable Server Pooling (RSerPool). We have shown via simulations that RSerPool, with the correct usage and configuration of pool member selection policies, can accomplish the cloud/edge setup resource selection task with a small overhead.

Intelligent proactive fault tolerance at the edge through resource usage prediction

The proliferation of demanding applications and edge computing establishes the need for efficient management of the underlying computing infrastructures, urging the providers to rethink their operational methods. In this paper, we propose an Intelligent Proactive Fault Tolerance (IPFT) method that leverages the edge resource usage predictions through Recurrent Neural Networks (RNNs). More specifically, we focus on the process faults, which are related with the inability of the infrastructure to provide Quality of Service (QoS) in acceptable ranges due to the lack of processing power. In order to tackle this challenge we propose a composite deep learning architecture that predicts the resource usage metrics of the edge nodes and triggers proactive node replications and task migration. Taking also into consideration that the edge computing infrastructure is also highly dynamic and heterogeneous, we propose an innovative Hybrid Bayesian Evolution Strategy (HBES) algorithm for automated adaptation of the resource usage models. The proposed resource usage prediction mechanism has been experimentally evaluated and compared with other state of the art methods with significant improvements in terms of Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE). Additionally, the IPFT mechanism that leverages the resource usage predictions has been evaluated in an extensive simulation in CloudSim Plus and the results show significant improvement compared to the reactive fault tolerance method in terms of reliability and maintainability.

Towards fairness and QoE-based edge allocation for multiplayer virtual reality applications in edge computing

Edge computing has emerged as the next big thing in distributed computing, by extending the cloud paradigm and offering efficient ways to engage with latency-intolerant applications, such as Virtual Reality (VR) multiplayer games. In edge computing, the service providers can benefit from existing cellular infrastructure to deploy services on edge servers that reside in close proximity to the users. Given the limited available budget for edge resource investment, one fundamental problem that manifests is the discovery of a prudent edge allocation strategy, that will efficiently prescribe which users are assigned to which edge servers, in order to tackle application-specific requirements, like minimizing system deployment costs. In this paper, considering the frequent interactions and view inconsistencies occurring among multiple users immersed in the same VR game, we address the problem from the users' perspective, focusing on improving their edge admission rate, resource provisioning and overall fairness, in order to subsequently maximize the average Quality of Experience (QoE). We call this the "Fairness and QoE-Based Edge Allocation" (FQEA) problem, formally formulating its properties and theoretically proving its complexity. However, discovering optimal solutions to the NP-hard FQEA in large-scale VR scenarios is challenging. Hence, we propose FQEA-H, a heuristic algorithm to generate allocation strategies in reasonable time. Comprehensive simulations, conducted on a real-world topological trace, demonstrate how FQEA-H can tackle the problem effectively, generally outperforming both the baseline and state-of-the-art alternatives.

Service Function Chaining in MEC: A Mean-Field Game and Reinforcement Learning Approach

Multiaccess edge computing (MEC) and network virtualization technologies are important enablers for fifth generation networks to deliver diverse services. Services are often provided as fully connected virtual network functions (VNFs), through service function chaining (SFC). However, the problem of allocating SFC resources at the edge still faces many challenges related to the way VNFs are placed, chained, and scheduled. In this article, to solve these problems, we propose a game theory-based approach with the objective to reduce service latency in the context of SFC at the edge. The problem of allocating SFC resources can be divided into two subproblems: 1) the VNF placement and routing subproblem, and 2) the VNF scheduling subproblem. For the former subproblem, we formulate it as a mean-field game in which VNFs are contending over edge resources aiming at reducing the resource consumption of MEC nodes and reducing latency for users. We also propose a reinforcement learning-based technique, where the Ishikawa--Mann learning algorithm is used. For the later subproblem, we formulate it as a matching game between VFNs and edge resources to find the execution order of the VNFs while reducing the latency. To efficiently solve it, we propose a modified version of the many-to-one deferred acceptance algorithm (DAA), called the enhanced multistep DAA. To illustrate the performance of the proposed approaches, we perform extensive simulations. The results show that the approaches achieve up to 40% less resource consumption, and up to 38% less latency than the benchmarked state-of-the-art methods.

Joint Application Placement and Request Routing Optimization for Dynamic Edge Computing Service Management

As mobile edge computing (MEC) hosting applications at the network edge with limited capacities, service providers are facing the new challenge of how to make full use of the scarce edge resources to maximize the system performance. Accommodating this challenge requires careful application placement and request routing to coordinate diverse MEC nodes. However, frequent application re-placement would greatly increase the system reconfiguration cost, indicating a performance-cost trade-off. In response, in this paper, we study the problem of joint optimization on application placement and request routing to maximize the system performance, under a long-term budget of the application reconfiguration cost. Solving this problem is non-trivial since the long-term budget is coupled with the future system states (e.g., user request arrivals) that are typically unpredictable. To address this challenge, we first advocate an approximated dynamic optimization framework to decompose the long-term optimization problem into a series of one-shot problems which do not require the future system states. Moreover, since the decomposed problem is a mixed integer linear program (MILP) which is proven to be NP-hard, we then devise an efficient dependent rounding based approximation algorithm, which can achieve the near-optimal performance in a fast manner. Both rigorous theoretical analysis and extensive trace-driven evaluations demonstrate the proposed framework can achieve superior performance gain over existing schemes.

Service-Oriented Resource Allocation for Blockchain-Empowered Mobile Edge Computing

Integrating dense small cell (DSC) networks with mobile edge computing is employed by 5G to tackle the contradiction between the computation limitations of user equipment (UE) and the stringent latency requirement of services. This paper investigates the service-oriented edge resource allocation problem in DSC networks, determining where to deploy the service entity, how many service entities should be deployed at each edge cloud, and how to assign the UEs to service entities. The problem is challenging for the following three aspects: 1) Service entity deployment and UE assignment are highly coupled. 2) Due to the overlap of coverage regions of densely deployed small cells, the allocation mechanism of different base stations has mutual effects on the overall service performance. 3) Considering the limited resources of edge clouds, it is a thorny problem to encourage edge clouds to cache and share service startup images. We devote the following efforts to tackle the problem under these challenges. First, we explore blockchain’s decentralized, traceable, and secure characteristics, and propose a scheme to encourage image sharing in mobile edge computing. Second, we formulate the service-oriented edge resource allocation as mixed integer non-linear programming. Third, towards the target of reducing the computational complexity, we decouple UE assignment from service entity deployment and solve it through Gibbs sampling. Moreover, the power of Lyapunov optimization and convex optimization is incorporated to reduce the long-term power consumption and budget. Experiment results demonstrate the superiority of our approach over current notable solutions.

Grid Edge Visibility: Gaps and a road map

Behind-The-Meter (BTM) resources at the grid edge are rapidly becoming an important component of the electric grid, requiring a substantial reconfiguration of traditional grid practices, such as policy changes, market redesign, and infrastructure upgrades. This adjustment is challenged by the fact that, by definition, grid edge elements are not easily observable by grid control entities. Increasing the visibility of these resources is therefore an important goal, one that is experiencing much research and discussion by various power system stakeholders. For example, policy makers are analyzing the tradeoffs of using grid edge meters to impose charges on grid edge electricity generation. System operators, such as the Midcontinent Independent System Operator (MISO) in the United States, can identify visibility information on the electrical location and the size of the grid edge resources as a critical consideration across the transmission-and-distribution (T&D) spectrum. This article summarizes the challenges and needs of grid entities resulting from the introduction of grid edge resources as well the gaps in the extant grid edge visibility frameworks.

Adaptive Federated Deep Reinforcement Learning for Proactive Content Caching in Edge Computing

With the aggravation of data explosion and backhaul loads on 5 G edge network, it is difficult for traditional centralized cloud to meet the low latency requirements for content access. The federated learning ( <u>F</u> L)-based <u>p</u> roactive content <u>c</u> aching (FPC) can alleviate the matter by placing content in local cache to achieve fast and repetitive data access while protecting the users’ privacy. However, due to the non-independent and identically distributed (Non-IID) data across the clients and limited edge resources, it is unrealistic for FL to aggregate all participated devices in parallel for model update and adopt the fixed iteration frequency in local training process. To address this issue, we propose a distributed resources-efficient FPC policy to improve the content caching efficiency and reduce the resources consumption. Through theoretical analysis, we first formulate the FPC problem into a stacked autoencoders (SAE) model loss minimization problem while satisfying resources constraint. We then propose an adaptive FPC (AFPC) algorithm combined deep reinforcement learning (DRL) consisting of two mechanisms of client selection and local iterations number decision. Next, we show that when training data are Non-IID, aggregating the model parameters of all participated devices may be not an optimal strategy to improve the FL-based content caching efficiency, and it is more meaningful to adopt adaptive local iteration frequency when resources are limited. Finally, experimental results in three real datasets demonstrate that AFPC can effectively improve cache efficiency up to 38.4 <inline-formula><tex-math notation="LaTeX">$\%$</tex-math></inline-formula> and 6.84 <inline-formula><tex-math notation="LaTeX">$\%$</tex-math></inline-formula> , and save resources up to 47.4 <inline-formula><tex-math notation="LaTeX">$\%$</tex-math></inline-formula> and 35.6 <inline-formula><tex-math notation="LaTeX">$\%$</tex-math></inline-formula> , respectively, compared with traditional multi-armed bandit (MAB)-based and FL-based algorithms.

HiTDL: High-Throughput Deep Learning Inference at the Hybrid Mobile Edge

Deep neural networks (DNNs) have become a critical component for inference in modern mobile applications, but the efficient provisioning of DNNs is non-trivial. Existing mobile- and server-based approaches compromise either the inference accuracy or latency. Instead, a hybrid approach can reap the benefits of the two by splitting the DNN at an appropriate layer and running the two parts separately on the mobile and the server respectively. Nevertheless, the DNN throughput in the hybrid approach has not been carefully examined, which is particularly important for edge servers where limited compute resources are shared among multiple DNNs. This article presents HiTDL, a runtime framework for managing multiple DNNs provisioned following the hybrid approach at the edge. HiTDL's mission is to improve edge resource efficiency by optimizing the combined throughput of all co-located DNNs, while still guaranteeing their SLAs. To this end, HiTDL first builds comprehensive performance models for DNN inference latency and throughout with respect to multiple factors including resource availability, DNN partition plan, and cross-DNN interference. HiTDL then uses these models to generate a set of candidate partition plans with SLA guarantees for each DNN. Finally, HiTDL makes global throughput-optimal resource allocation decisions by selecting partition plans from the candidate set for each DNN via solving a fairness-aware multiple-choice knapsack problem. Experimental results based on a prototype implementation show that HiTDL improves the overall throughput of the edge by <inline-formula><tex-math notation="LaTeX">$4.3\times$</tex-math></inline-formula> compared with the state-of-the-art.

Edge resource slicing approaches for latency optimization in AI-edge orchestration

Edge resource slicing approaches for latency optimization in AI-edge orchestration

Scheduling Service Updates: A Multi-arm Bandit Approach

Software as a Service (SaaS) instances often use edge resources to serve their customers. The version of the service hosted at the edge needs to be periodically updated to maximize the utility derived by the customers. We focus on scheduling updates in the setting where the utility derived from a version is an unknown decreasing function of the time elapsed since the version was created. We map the scheduling problem to a multi-arm bandit and propose an update policy. We characterize its performance and compare it with the fundamental limit on the performance of any online policy.

Технології реалізації штучного інтелекту як сервісу на основі апаратних прискорювачів

The subject of study in this article is modern technologies, tools and methods of building AI systems as a service using FPGA as a platform. The goal is to analyze modern technologies and tools used to develop FPGA-based projects for systems that implement artificial intelligence as a service and to prepare a practical AI service prototype. Task: to analyze the evolution of changes in the products of leading manufacturers of programmable logic devices and experimental and practical examples of the implementation of the paradigm of continuous reprogramming of programmable logic; analyze the dynamics of changes in the development environment of programmable logic systems for AI; analyze the essential elements of building projects for AI systems using programmable logic. According to the tasks, the following results were obtained. The area of application of hardware implementation of artificial intelligence for on-board and embedded systems including airspace industry, smart cars and medical systems is analyzed. The process of programming FPGA accelerators for AI projects is analyzed. The analysis of the capabilities of FPGA with HBM for building projects that require enough of high speed memory is performed. Description languages, frameworks, the hierarchy of tools for building of hardware accelerators for AI projects are analyzed in detail. The stages of prototyping of AI projects using new FPGA development tools and basic DPU blocks are analyzed. The parameters of the DPU blocks were analyzed. Practical steps for building such systems are offered. The practical recommendations for optimizing the neural network for FPGA implementation are given. The stages of neural network optimization are provided. The proposed steps include pruning of branches with low priority and the use of fixed point computations with custom range based on the requirements of an exact neural network. Based on these solutions, a practical case of AI service was prepared, trained and tested. Conclusions. The main contribution of this study is that, based on the proposed ideas and solutions, the next steps to create heterogeneous systems based on the combination of three elements are clear: AI as a service, FPGA accelerators as a technology for improving performance, reliability and security, and cloud or Edge resources to create FPGA infrastructure and AI as service. The development of this methodological and technological basis is the direction of further R&amp;D.