Predição do desempenho de programas paralelos por simulação do grafo de execução

AUTOR(ES)
DATA DE PUBLICAÇÃO

1997

RESUMO

The process of writing a new code for a parallel programming environment demands from its designer a lot of attention on the performance achieved by the pair program-machine. The high processing costs justify the efforts to reduce processing time and costs, which leads to the problem of defining performance metrics and approaches to measure the system performance. Several performance analysis and prediction tools are available to help designers in such a task. With those tools the designer can locate critical points in the code and look for solutions to improve the program s performance. Unfortunately, most of those tools make so many approximations in the parallel environment model, that make themselves somewhat inaccurate. Moreover method of measurement usually adopted get performance data through an instrumented program s run. This approach affects the measures since additional code is inserted into the program under analysis. This work introduces a new method to make such measurements without including additional code nor requiring runs on the target parallel machine. The proposed approach is to rewrite the executable code into an execution graph, which is a directed graph that keeps the information about the processing time of each machine instruction included in the code under analysis... Note: The complete abstract is available with the full electronic digital thesis or dissertations

ASSUNTO(S)

desempenho programas de computador processamento paralelo (computadores) simulação (computadores)

ACESSO AO ARTIGO

Documentos Relacionados