Ye-jin Kwon, Young-bom Park, Jeong-cheol Lee, Journal of Internet Computing and Services, Vol. 26, No. 2, pp. 11-21, Apr. 2025
10.7472/jksii.2025.26.2.11, Full Text:  HTML
Keywords: HPC Simulation, Heterogeneous Cluster, Slurm, Kubernetes, Container Image, EDISON Platform

Abstract

This paper, we design, develop, and test a web platform for executing high-performance computing (HPC) simulations in heterogeneous cluster environments. We builds a container-based simulation execution environment that accommodates the application requirements, libraries, and execution processes necessary for simulations in various computational science and engineering fields. Additionally, we proposed platform leverages containerization to ensure the independence of simulation applications, facilitating smooth execution, data preprocessing, and visualization. The platform that we proposed integrates Slurm and Kubernetes cluster frameworks to enable efficient resource management, dynamic scheduling, and task distribution, while offering easy simulation management through a user-friendly web interface. We performance comparison experiments between Slurm and Kubernetes clusters, using the EDISON platform. Accordingly, we provide key analysis results on execution time and resource utilization for various simulation applications. As a result, we analysis that Slurm demonstrated superior performance for applications requiring large-scale data and high-performance resources, while Kubernetes proved to be more adaptable for small-scale and dynamic workloads. Our purpose of the first step of this study, lays the groundwork for analyzing execution environment optimization algorithms and resource allocation strategies for heterogeneous cluster-based HPC simulations. As future research we will focus on improving the design of interconnected simulation workflows and exploring advanced scheduling techniques for heterogeneous clusters.

Statistics
Cite this article