Elsevier

Computer Networks

Volume 189, 22 April 2021, 107916
Computer Networks

Cooperative service caching and computation offloading in multi-access edge computing

https://doi.org/10.1016/j.comnet.2021.107916Get rights and content

Abstract

Multi-access edge computing (MEC) as an emerging and promising paradigm can alleviate the physical resource bottlenecks for smart mobile devices, involving storage and processing capacities. In the MEC system, the traffic load and the quality of service (QoS) can be improved through service caching. However, due to the highly coupled relationship between service caching and offloading decisions, it is extremely challenging to flexibly configure edge service cache within limited edge storage capacity to improve system performance. In this paper, we aim to minimize the average task execution time in the edge system by considering the heterogeneity of task requests, the pre-storage of the application data, and the cooperation of the base stations. Firstly, we formulate the problem of joint computation offloading, service caching, and resource allocation as a Mixed Integer Non-Linear Programming (MINLP) problem, which is difficult to solve because of the coupling relationship between optimization variables. Then we solve the MINLP problem by the decomposition theory based on Generalized Benders Decomposition. Moreover, we develop an efficient algorithm of cooperative service caching and computation offloading, called GenCOSCO, to improve QoS while reducing computation complexity. In particular, for special cases when the service cache configuration is fixed, the FixSC algorithm is proposed to derive the offloading strategy by cache replacement. Finally, numerous simulation experiments indicate that our proposed scheme can significantly reduce the average time consumption of task execution.

Introduction

The rapid development of mobile cloud computing and internet of things has driven the explosive of emerging computation-intensive and latency-sensitive applications, such as augmented/virtual reality (AR/VR), natural language processing, face recognition (FR) and so on [1], [2]. The widespread use of these applications has led to the fact that smart mobile devices (SMDs) with limited computation and storage resources are no longer sufficient as the primary operating platform. To overcome these limitations, putting computation tasks on the remote cloud has been adopted to be an effective way of data processing due to its abundant computation resources [1]. Offloading excessive computation tasks to the remote cloud inevitably introduces severe communication delays due to long transmission distance. One feasible method is to drop the computing process to the network’s logical edge to reduce communication delay. Therefore, the multi-access edge computing (MEC) as a promising technology, can bring data analysis and information storage closer to the data source generated by user devices.

An MEC system primarily consists of edge servers endowed with cloud-like computing and storage capabilities, and the base station (BS) deployed near the SMDs. The MEC server is enabled to serve as an attractive substitute of the remote cloud for delay-sensitive offloading tasks. However, the resources of the MEC servers are more limited than remote cloud. When the server is overloaded, the system would further offload the extra tasks to the cloud. This constitutes a hierarchical offloading network structure, which naturally causes the question of where the offloading task is performed. However, previous researches on the central topic of computation offloading often ignore the heterogeneity of mobile services and do not consider how these services are cached to BS.

To fulfill the requirements of latency-sensitive tasks and the actual demand of the applications that require pre-storage some non-trivial data (e.g., AR and interactive online gaming). The concept of service caching (i.e., service placement) is proposed to further optimize the quality of service (QoS) of the system. The MEC server involves caching application services and their corresponding data or libraries so that computing tasks can be performed. In general, these services can be loaded during the cold start initialization of computing platform, which mainly consists of initializing cloud functions, necessary service environments, and specific user code [3]. However, the MEC server can only provide a set of services for subscribers since its storage capacity is much smaller than that of remote cloud. Therefore, a significant design goal is to cache the necessary service in the systems selectively. The above issue is further complicated by the nature of application service heterogeneity and the diversity of SMDs requirements.

In addition, the rapid development of 5G technology constitutes a complex multi-cellular network environment. It enables multiple MEC servers to cooperate and offload tasks through the backhaul network, and to achieve load balancing and resource sharing within the system. Specifically, with the supports of the cooperative caching and computation resources allocation, the overall system operating time can be improved by mapping users to BS that are more suitable for their service demands and latency requirements, thereby reducing the link load caused by retrieving services from the remote cloud. Therefore, the effective use of MEC resources is necessary to guarantee its foreseeable benefits, which are strictly related to solving the following challenges: (1) computation offloading, which needs to determine the destination server based on the system conditions; (2) services caching, which requires adaptive adjustment according to user service requests; (3) resource allocation, which requires a reasonable allocation of BS resources to achieve system goals.

To address these challenges, we design a novel scheme of joint computation offloading, collaborative service caching, and resource allocation to reduce the link traffic load and user-perceived latency. Compared with the previous works, our contributions in this work can be summarized as follows:

  • We propose a mutually beneficial cooperation framework to minimize the average execution latency in the MEC system, where network conditions are taken into consideration to select the target BS reasonably. Meanwhile, we formulate the problem of Cooperative Service Caching and Computation Offloading (COSCO) as a Mixed Integer Non-linear Programming (MINLP) that jointly optimizes computation and storage resources.

  • To reduce computational complexity, we employ an auxiliary variable to transform the original optimization problem into a decomposable problem, and then utilize the modified Generalized Benders Decomposition (GBD) method to obtain the decision strategy of the system.

  • We further propose two algorithms, i.e., FixSC algorithm and GenCOSCO algorithm, to solve the COSCO problem. The FixSC algorithm is used to generate the offloading decision under the fixed service cache configuration, which applies the cache replacement to implement dynamic service caching. Then, for the general scenarios, the GBD-based GenCOSCO algorithm is employed to determine offloading and resource allocation strategy.

  • We conduct simulations to evaluate the performance of our proposed algorithms. The numerical result indicates that our algorithms can effectively reduce the delay cost and efficiently utilize computation and storage resources.

The remainder of this paper is organized as follows. In Section 2, the review of related work is presented. In Section 3, we outline the system model and present the problem formulation. Then in Section 4, we transform the original problem, and the general benders decomposition method is exploited to solve the optimization problem. In Section 5, we describe our algorithms for the COSCO problem. In Section 6, we implement the scheme on a simulation testbed and provide the experiment results, followed by the conclusion in Section 7.

Section snippets

Related works

There have been some existing works focusing on computation offloading and resource allocation to release the computation burden of SMDs in the MEC system. The previous studies on computation offloading [4], [5], [6], [7], [8], [9], aim to determine whether or how much task data to be offloaded. According to the characteristic of computation tasks, computation offloading is either based on the principle of partial offloading [4], whose tasks can be divided into multiple components for

System model and problem formulation

In this section, we will introduce the system model. Due to the dense deployment of BSs in 5G environment, we assume that the network is divided into multiple disjoint regions, such that edge servers within the same region M will form a shared resource pool. Consequently, SMDs in region M have an accessible set of edge servers in multi-hop manner. Therefore, we consider an edge-cloud system in the region M shown in Fig. 1 with K BSs equipped with MEC servers endowed with computing and storage

Problem transformation and solution

In this section, we focus on solving the joint optimization problem, aiming to minimize system average execution latency. We decouple the continuous and discrete variables using the Modified General Benders Decomposition (MGBD) method.

Algorithms

In this section, we represent our algorithms to solve COSCO problem. From the perspective of the mobile edge service operator, frequent updates of the intra-network service cache will result in inevitable operating costs. Thus we assume that the service cache policy is updated regularly for the system and keeps unchanged for a certain period, but the computation offloading and resource allocation strategies are adjusted instantaneously with the user task status.

Performance evaluation

In this section, we consider an MEC network covered with 500 m × 500 m region served by 2 BSs. The coverage radius of all BSs is set to be 200m, where the SMDs are randomly scattered in the region. Each SMD has some computation-intensive tasks that need to be executed. We set the input data size of each task to be randomly distributed within 2001000 KB, and its corresponding number of CPU cycles is distributed between 0.41.2 Gcycles. The service needs to pre-cache a certain size of data, and

Conclusion and future works

In this paper, we study the problem of joint optimization on computation offloading, services caching, and resource allocation for MEC-enabled cellular networks, aiming to improve the system QoS by reducing the communication and computation delay. We first formulate the joint optimization problem as an MINLP model. To reduce high computational complexity, we employ the MGBD method and present a GBD-based GenCOSCO algorithm to determine the computation offloading and resource allocation

CRediT authorship contribution statement

Shijie Zhong: Conceptualization, Methodology, Algorithm presentation, Writing - original draft. Songtao Guo: Idea provision, Supervision, Writing - review & editing. Hongyan Yu: Data curation, Algorithm implementation. Quyuan Wang: Performance evaluation, Algorithm validation.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgments

This work was supported by the National Natural Science Foundation of China (No. 61772432, 617724333), Natural Science Key Foundation of Chongqing (cstc2020jcyj-zdxmX0026), Fundamental Research Funds for the Central Universities (2020CDCGJSJ071, 2020CDCGJSJ038, 2019CDYGZD004), and Zhejiang Lab (NO. 2021LC0AB01).

Shijie Zhong received the BS degree in communication engineering from Southwest University, Chongqing, China, in 2018. He is currently pursuing the MS degree in the college of Electrical Information and Engineering, Southwest University, Chongqing, China. His current research interests include edge computing, resource allocation, convex optimization theory and its application.

References (35)

  • ChenL. et al.

    Collaborative service placement for edge computing in dense small cell networks

    IEEE Trans. Mob. Comput.

    (2019)
  • BagajewiczM. et al.

    On the generalized benders decomposition

    Comput. Chem. Eng.

    (1991)
  • ShiW. et al.

    Edge computing: Vision and challenges

    IEEE Internet Things J.

    (2016)
  • AlameddineH.A. et al.

    Dynamic task offloading and scheduling for low-latency iot services in multi-access edge computing

    IEEE J. Sel. Areas Commun.

    (2019)
  • E. Jonas, J. Schleier-Smith, V. Sreekanti, C.-C. Tsai, A. Khandelwal, Q. Pu, V. Shankar, J. Carreira, K. Krauth, N....
  • SaleemU. et al.

    Latency minimization for d2d-enabled partial computation offloading in mobile edge computing

    IEEE Trans. Veh. Technol.

    (2020)
  • J. Liu, Y. Mao, J. Zhang, K.B. Letaief, Delay-optimal computation task scheduling for mobile-edge computing systems,...
  • YangL. et al.

    Mobile edge computing empowered energy efficient task offloading in 5g

    IEEE Trans. Veh. Technol.

    (2018)
  • Y. Geng, Y. Yang, G. Cao, Energy-efficient computation offloading for multicore-based mobile devices, in: IEEE INFOCOM...
  • ChenW. et al.

    Multi-user multi-task computation offloading in green mobile edge cloud computing

    IEEE Trans. Serv. Comput.

    (2019)
  • NingZ. et al.

    A cooperative partial computation offloading scheme for mobile edge computing enabled internet of things

    IEEE Internet Things J.

    (2019)
  • YouC. et al.

    Energy-efficient resource allocation for mobile-edge computation offloading

    IEEE Trans. Wireless Commun.

    (2017)
  • YouC. et al.

    Asynchronous mobile-edge computation offloading: Energy-efficient resource management

    IEEE Trans. Wireless Commun.

    (2018)
  • XingH. et al.

    Joint task assignment and resource allocation for d2d-enabled mobile-edge computing

    IEEE Trans. Commun.

    (2019)
  • TranT.X. et al.

    Joint task offloading and resource allocation for multi-server mobile-edge computing networks

    IEEE Trans. Veh. Technol.

    (2018)
  • LiuY. et al.

    Distributed resource allocation and computation offloading in fog and cloud networks with non-orthogonal multiple access

    IEEE Trans. Veh. Technol.

    (2018)
  • BiS. et al.

    Computation rate maximization for wireless powered mobile-edge computing with binary computation offloading

    IEEE Trans. Wireless Commun.

    (2018)
  • Cited by (27)

    View all citing articles on Scopus

    Shijie Zhong received the BS degree in communication engineering from Southwest University, Chongqing, China, in 2018. He is currently pursuing the MS degree in the college of Electrical Information and Engineering, Southwest University, Chongqing, China. His current research interests include edge computing, resource allocation, convex optimization theory and its application.

    Songtao Guo received the BS, MS, and Ph.D. degrees in computer software and theory from Chongqing University, Chongqing, China, in 1999, 2003, and 2008, respectively. He was a professor from 2011 to 2012 at Chongqing University and a professor from 2012 to 2018 at Southwest University. He is currently a full professor at Chongqing University, China. He was a senior research associate at the City University of Hong Kong from 2010 to 2011, and a visiting scholar at Stony Brook University, New York, from May 2011 to May 2012. His research interests include mobile edge computing, mobile cloud computing and parallel and distributed computing. He has published more than 100 scientific papers in leading refereed journals and conferences. He has received many research grants as a principal investigator from the National Science Foundation of China and Chongqing and the Postdoctoral Science Foundation of China.

    Hongyan Yu received the Ph.D. degree at the College of Electrical Information and Engineering, Southwest University, Chongqing, China, in 2017. He is currently an associate professor at Chongqing Three Gorges University, China. His current research interests include physical layer security, mobile edge computing, and convex optimization theory and its application.

    Quyuan Wang received the B.S. degree in communication engineering from Southwest University, Chongqing, China, in 2012. He is currently pursuing the Ph.D. degree in the college of Electrical Information and Engineering, Southwest University, Chongqing, China. His current research interests include game theory, edge computing, auction theory, convex optimization theory and its application.

    View full text