International Journal of Electrical and Computer Engineering (IJECE) Vol. 10, No. 2, April 2020, pp. 1507~1514 ISSN: 2088-8708, DOI: 10.11591/ijece.v10i2.pp1507-1514  1507 Journal homepage: http://ijece.iaescore.com/index.php/IJECE Performance analysis of container-based networking Solutions for high-performance computing cloud Sang Boem Lim 1 , Joon Woo 2 , Guohua Li 3 1 Department of Smart ICT Convergence, Konkuk University, Korea 2, 3 National Institute of Supercomputing and Networking, Korea Institute of Science and Technology Information, Korea Article Info ABSTRACT Article history: Received Jul 4, 2019 Revised Oct 11, 2019 Accepted Oct 20, 2019 Recently, cloud service providers have been gradually changing from virtual machine-based cloud infrastructures to container-based cloud-native infrastructures that consider performance and workload-management issues. Several data network performance issues for virtual instances have arisen, and various networking solutions have been newly developed or utilized. In this paper, we propose a solution suitable for a high-performance computing (HPC) cloud through a performance comparison analysis of container-based networking solutions. We constructed a supercomputer-based test-bed cluster to evaluate the serviceability by executing HPC jobs. Keywords: Cloud computing Container-based networking HPC Network Performance analysis Copyright © 2020 Institute of Advanced Engineering and Science. All rights reserved. Corresponding Author: Guohua Li, National Institute of Supercomputing and Networking, Korea Institute of Science and Technology Information, 245 Daehak-ro, Yuseong-gu, Daejeon, 34141, Korea. Email: ghlee@kisti.re.kr 1. INTRODUCTION Traditionally, high-performance computing (HPC) was used mainly by natural science areas like weather forecasting, molecular biology, and space exploitation. HPC becomes very demanding technology due to the importance of big data analysis, that cannot be processed with the traditional computing environment [1]. A wide range of computer architecture is, also, required to process big data. One solution to these requirements is to add cloud capability to the HPC environment. Cloud computing can easily adart rapid changes of hardware and sofrtware techologies. By using cloud computing, most of the users also can reduce analysis time and cost of hardware and software [2]. The design of networks in a cloud infrastructure configuration is usually divided into public, management, and guest networks [3]. A guest network is capable of data communication between virtual instances which are running on one host, multiple hosts, or across the different subnets [4-6]. The Docker [7] container platform is an open-source container management project launched by Docker Inc. This is a lightweight container technology that bundles and runs the service operating environment. When configuring a cloud environment based on the Docker container, an orchestration software such as Kubernetes or Docker Swarm is needed to effectively manage and efficiently allocate the resources required for containers [8-10]. The goal of this paper is to evaluate a suitable container-based network solution for HPC cloud by performing benchmark tests using Message Passing Interface (MPI) benchmark suite. We have tested several networking solutions using Kubernetes which has an excellent auto-recovery capability. This study includes a result of the performance tests on network bandwidth of various cluster network configurations and an evaluation of the HPC serviceability of the bare-metal and container. Additionally, a summary of the evaluation result and recommendation on container-based network solution is provided.