HPC-SBC: um esforço experimental para avaliar o armazenamento em configurações de computação de alto desempenho usando uma abordagem contextual

Abstract

High-performance environments often face input and output (I/O) bottlenecks. One factor is the gap between processing power and storage performance. This is because storage technologies evolve at a slower speed than processing, contributing to the increase of I/O performance problems. Thus, increasingly relevant contributions are being made to mitigate I/O gaps. This dissertation, therefore, presents an experimental effort to evaluate the performance of the secondary memory of computing nodes in high-performance environments, using the Grid’5000 environment as an example. In this way, information about the quantity, type and performance of node storage devices in parallel distributed environments is obtained in a more technical and scientific way. In addition to this, a proposal was developed to analyze the storage of these high-performance environments, which was called HPC-SBC (HPC Storage Benchmark Context). This computational approach can overcome the limitations observed in the tools available in the Grid’5000 computing environment, by providing differentiated information about the performance of the secondary memory devices available in each node in this environment. This tool allows you to analyze the transfer rate of each device found and also provide users with automated information about the performance of storage devices in the cluster context. This contextual approach provided initial results that indicate the success of the research, by pointing out among the nodes analyzed in the high-performance processing environment (High Performance Computing - HPC), those storage devices with the best performance.