A Novel Hybrid Model for Task Dependent Scheduling in Container-based Edge Computing Tingting Lu*,Fanping Zeng*,Guozhu Chen*,Wenjuan Shu*,Jingfei Shen*and Weikang Zhang* *School of Computer Science and Technology.University of Science and Technology of China,Hefei,Anhui,China 120c TAnhui Province Key Lab of Software in Computing and Communication,Hefei,Anhui,PR China ftingtlu,chengz18,shuwj.ericjeff,buttman@mail.ustc.edu.cn,billzeng@ustc.edu.cn 330085 Abstract-In traditional edge computing,the task from the configured in the edge server.That is to say,if a task is planned Internet of Things (IoT)is usually offloaded to edge server.It on an edge server that has no corresponding container,the will be uploaded to the remote cloud if the edge server cannot server can configure the container which is downloaded from process it.A task can be processed on the server,only if the server has configured the corresponding function program.However, the remote cloud.After the container has been configured,their each edge server can only configure a small number of functions server can process the task.Therefore,it is a key challenge to due to the limited computing,storage,and bandwidth resources. design a task scheduling algorithm that considers the container Moreover,modern tasks from IoT devices become more and more configuration in EC. diverse,which are also accompanied by complex dependencies Second,the application (also called a job)from IoT de- 3331 It increases the processing time overhead to the task processed in remote cloud due to huge transmission delay.In this paper,we vice consists of some tasks that have complex dependencies design a container-based edge computing system,where a task each of which needs a suitable container.Specifically,the can be executed on a server only if the server has configured the communication among servers will occur between dependent corresponding container,if not the server can fetch it from other 10000.1/0001007820-1-86 tasks when they are allocated on different servers respectively. edge servers or remote cloud.Based on the system,we propose a Furthermore,because of the dependency relationship,the start novel hybrid model,called CBASGA,with the aim to minimize the job complete time,which combines Chaos-based Beetle time of task will be limited by its direct predecessors.In other Antennae Search and Genetic Algorithm.Our experimental words,a task can only be executed after all of its direct prede- results show that the designed system reduces the average job cessors have been completed.So how to deal with the complex completion time by 4.2%compared with the comparison system, inter-task dependency is another challenge in EC.There have and CBASGA reduces the average job completion time by at been many researches on task scheduling and related server least 21.7%compared with baselines. Index Terms-Task scheduling,Container configuration,Edge configuration (e.g.,[6,71).But they consider either scheduling computing the whole job individually [7,8]or configuring the related (sdo servers independently [9]. I.INTRODUCTION In this paper,we first propose a container-based edge system in which containers can be transferred between edge servers. Edge computing (EC)provides resource services (e.g.,com- The edge servers have limited resources and each server puting,storage,bandwidth,etc)to nearby users by deploying has a different performance.We then propose an efficient small servers (called edge server)in the edge of the network approximation algorithm,named CBASGA.The algorithm [1].Nearby users can offload their tasks or jobs to edge combines Chaos-based Beetle Antennae Search (CBAS)and server for processing,so that the data transmission time can Genetic Algorithm (GA)in addition to heuristic initialization be greatly reduced. to optimize the tradeoff between task dependency and con- Due to that the application pre-configured by the developer tainer configuration.Finally,we adopt a cluster trace from in edge server is limited,the edge server can only process a Alibaba to evaluate CBASGA.The simulation results show few tasks.It can not take full advantage of edge computing. that the designed system is efficient to process dependent tasks So some scholars have proposed containerizing the applica- and CBASGA has excellent performance. tions required for IoT devices,which can be automatically The remainder of the paper is organized as follows.Section obtained from the cloud when there is no corresponding II expounds our edge system model.The task scheduling containerized application for processing task[2].Furthermore, and correlation allocation algorithms are shown in section III. Martin Fowler and James Lewis jointly propose the concept Section IV presents the performance evaluation for CBASGA. of microservices and define a microservice as a small service We conclude this paper in section V. composed of a single application,which can be deployed, II.SYSTEM MODEL AND PROBLEM FORMULATION scaled,and tested independently [3].Many works have applied the idea of microservices to edge computing [4,5].But there A.Edge System Model are still many challenges in applying microservices to EC. The edge system consists of K heterogeneous edge servers First,the storage and computing resources of edge services and one remote cloud,denoted as S={so,s1,...,sK,where are limited.Only a few containers (i.e.,microservices)can be so depicts the remote cloud.We use rkas the data transfer 978-1-7281-9441-721/S31.0002021EEE Authorized licensed use limited to:University of Science Technology of China.Downloaded on July 14.2021 at 02:23:40 UTC from IEEE Xplore.Restrictions apply.A Novel Hybrid Model for Task Dependent Scheduling in Container-based Edge Computing Tingting Lu∗, Fanping Zeng∗†, Guozhu Chen∗, Wenjuan Shu∗, Jingfei Shen∗ and Weikang Zhang∗ ∗School of Computer Science and Technology, University of Science and Technology of China, Hefei, Anhui, China †Anhui Province Key Lab of Software in Computing and Communication, Hefei, Anhui, PR China {tingtlu, chengz18, shuwj, ericjeff, buttman}@mail.ustc.edu.cn, billzeng@ustc.edu.cn Abstract—In traditional edge computing, the task from the Internet of Things (IoT) is usually offloaded to edge server. It will be uploaded to the remote cloud if the edge server cannot process it. A task can be processed on the server, only if the server has configured the corresponding function program. However, each edge server can only configure a small number of functions due to the limited computing, storage, and bandwidth resources. Moreover, modern tasks from IoT devices become more and more diverse, which are also accompanied by complex dependencies. It increases the processing time overhead to the task processed in remote cloud due to huge transmission delay. In this paper, we design a container-based edge computing system, where a task can be executed on a server only if the server has configured the corresponding container, if not the server can fetch it from other edge servers or remote cloud. Based on the system, we propose a novel hybrid model, called CBASGA, with the aim to minimize the job complete time, which combines Chaos-based Beetle Antennae Search and Genetic Algorithm. Our experimental results show that the designed system reduces the average job completion time by 4.2% compared with the comparison system, and CBASGA reduces the average job completion time by at least 21.7% compared with baselines. Index Terms—Task scheduling, Container configuration, Edge computing I. INTRODUCTION Edge computing (EC) provides resource services (e.g., com￾puting, storage, bandwidth, etc) to nearby users by deploying small servers (called edge server) in the edge of the network [1]. Nearby users can offload their tasks or jobs to edge server for processing, so that the data transmission time can be greatly reduced. Due to that the application pre-configured by the developer in edge server is limited, the edge server can only process a few tasks. It can not take full advantage of edge computing. So some scholars have proposed containerizing the applica￾tions required for IoT devices, which can be automatically obtained from the cloud when there is no corresponding containerized application for processing task [2]. Furthermore, Martin Fowler and James Lewis jointly propose the concept of microservices and define a microservice as a small service composed of a single application, which can be deployed, scaled, and tested independently [3]. Many works have applied the idea of microservices to edge computing [4, 5]. But there are still many challenges in applying microservices to EC. First, the storage and computing resources of edge services are limited. Only a few containers (i.e., microservices) can be configured in the edge server. That is to say, if a task is planned on an edge server that has no corresponding container, the server can configure the container which is downloaded from the remote cloud. After the container has been configured, their server can process the task. Therefore, it is a key challenge to design a task scheduling algorithm that considers the container configuration in EC. Second, the application (also called a job) from IoT de￾vice consists of some tasks that have complex dependencies, each of which needs a suitable container. Specifically, the communication among servers will occur between dependent tasks when they are allocated on different servers respectively. Furthermore, because of the dependency relationship, the start time of task will be limited by its direct predecessors. In other words, a task can only be executed after all of its direct prede￾cessors have been completed. So how to deal with the complex inter-task dependency is another challenge in EC. There have been many researches on task scheduling and related server configuration (e.g., [6, 7]). But they consider either scheduling the whole job individually [7, 8] or configuring the related servers independently [9]. In this paper, we first propose a container-based edge system in which containers can be transferred between edge servers. The edge servers have limited resources and each server has a different performance. We then propose an efficient approximation algorithm, named CBASGA. The algorithm combines Chaos-based Beetle Antennae Search (CBAS) and Genetic Algorithm (GA) in addition to heuristic initialization to optimize the tradeoff between task dependency and con￾tainer configuration. Finally, we adopt a cluster trace from Alibaba to evaluate CBASGA. The simulation results show that the designed system is efficient to process dependent tasks and CBASGA has excellent performance. The remainder of the paper is organized as follows. Section II expounds our edge system model. The task scheduling and correlation allocation algorithms are shown in section III. Section IV presents the performance evaluation for CBASGA. We conclude this paper in section V. II. SYSTEM MODEL AND PROBLEM FORMULATION A. Edge System Model The edge system consists of K heterogeneous edge servers and one remote cloud, denoted as S = {s0, s1, ..., sK}, where s0 depicts the remote cloud. We use rk,k as the data transfer 978-1-7281-9441-7/21/$31.00 ©2021 IEEE 2021 IEEE International Conference on Communications Workshops (ICC Workshops) | 978-1-7281-9441-7/20/$31.00 ©2021 IEEE | DOI: 10.1109/ICCWorkshops50388.2021.9473877 Authorized licensed use limited to: University of Science & Technology of China. Downloaded on July 14,2021 at 02:23:40 UTC from IEEE Xplore. Restrictions apply