Análise de desempenho de algoritmos de escalonamento aplicados a um sistema gerenciador de recursos para execução de aplicação operacional de computação científica

Abstract

During the last years, the High Performance Computing (HPC) is one of the main research areas in computing, with the proposal to increase the performance of high performance environments and decrease the application time execution. With lot distributed computational resources, the Resource Management System (RMS) is the response to the manager of computational resources, job manager and user manager in HPC. For perform the management in HPC, the RMS have some essential components for its operation, between them, one of the most important is a job scheduling, that defines when and where each job will execute. Yet, the job scheduling choice for scientific environments that require results in a timely manner and high utilization of the computational resources still need more researches and experiments. In this study, I analyzed the aspects of the distributed environment computational when we used four job scheduling different applied in resource management system SLURM, in an environment that running a scientific application of ionosphere forecasting. We present the strong points and weak points of each one job scheduling algorithm developed applied in resource management system when exposed in adverse situation in an environment that performs a scientific application.