Classificação de um benchmark paralelo para arquiteturas multinúcleo

Abstract

This work aims to classify a set of thirteen applications used to measure performance and energy power consumption in multicore architectures. These applications are parallelized using four parallel programming interfaces: OpenMP, Pthreads, MPI-1 and MPI-2. They were developed in previous studies and classified according to some criteria, such as: number of operations performed in memory, communication rate and number of data exchange operations between threads/processes. In this work we perform a series of tests in these applications, aiming to classify them and show that they have enough characteristics to be used as a benchmark to evaluate the performance and energy consumption of different parallel programming interfaces in multicore architectures. The applications were classified according to: data structures, complexities, TLP, CPU usage and memory usage. The thirteen applications and their implementations with four PPIs cover the classification criteria in such a way that several scenarios can be evaluated, being efficient as benchmark for performance analysis and energy consumption of multicore architectures.