Design and development of a cloud-based HPC system for scientific computing

Abstract

The rapid development of research based on processing large amounts of data has increased the need for high-performance computing (HPC) solutions that combine power with flexibility, scalability and cost-effectiveness. Traditional on-premises HPC infrastructures face limitations related to scaling hardware resources, high maintenance costs, and inefficient use of computing power. In this context, cloud-based HPC systems are becoming a promising alternative, providing access to on-demand resources and supporting a wide range of scientific tasks.

This paper presents the architecture and implementation of a cloud-based HPC system focused on scientific computing. The proposed solution combines virtualized and containerized resources, high-speed interconnections, distributed storage, and automated resource management mechanisms. Special attention is paid to workload orchestration, scalability, fault tolerance, and efficient use of heterogeneous resources, including CPU and GPU. The system supports parallel computing models such as MPI and OpenMP, as well as workflows specific to scientific simulations, modeling, and data analysis.

The results of the experimental evaluation show that the proposed cloud-based HPC system provides performance comparable to traditional cluster solutions, while offering higher flexibility and scalability. The data obtained confirms that cloud technologies are able to effectively support high-performance computing tasks, reducing infrastructure costs and increasing accessibility for research organizations. The findings of the study contribute to the development of next-generation HPC platforms and provide practical recommendations for deploying scientific computing workloads in a cloud environment