Implantação e análise de desempenho de um cluster com processadores ARM e plataforma raspberry Pi

Abstract

With the recent advancements in High Performance Computing (HPC), it is possible to rapidly process high volumes of data, allowing accomplishments in several areas of knowledge. Although the HPC area has been observed as an area of complex infrastructure, in the last years, its been observed that the processing power of processors used in embedded systems, using the ARM architecture, has been increasing significantly. Furthermore, the acquisition costs and energy consumption are lower, when compared to processors of other platforms, thus allowing for the possibility of having HPC with smaller and more economical platforms, with lower maintenance cost and more accessible. In this merit, this masters thesis proposes the performance analysis of a low cost embedded cluster composed of processors using ARM architecture and the Raspberry Pi platform. This work analysis the impact of using MPICH-2 and OpenMPI libraries, running benchmark programs HPCC and HPL. The present work show results of performance and energy consumption of this cluster with these programs, proving that it is possible to use clusters of low cost embedded platforms with satisfactory speedups and energy consumption.