Análise de Desempenho de Programas Paralelos em Redes de Workstations
Abstract
The number of applications that demand high performance computing systems has increased in the last years. Traditionally, supercomputers have been used to run large applications. An attractive alternative to run those numerically intensive applications seems to be the use of a network of workstations. This paper analyzes the performance of parallel programs executing on a network of workstations. The paper also identifies conditions where the parallel processing outperforms the sequential execution. With the goal of increasing the performance at parallel processing, the paper proposes and analyzes the use of heterogeneous workstations to overcome the bottlenecks imposed by sequential fraction and communication existing in a parallel program.
References
Annamaria Benzoni, Vaidy S. Sunderam, Robert van de Geijn. "Matriz Factorization on a RISC Workstation Network", High Performance Computing II,1991,p. 207.
G. Fox, M. Johnson, G. Lyzenga, S. Otto, J. Salmon, D. Walker. "Solving Problems on Concurrent Processors", Prentice-Hall Internacional Editions, 1988.
F. A. Briggs, K. Hwang. "Computer Architecture and Parallel Processing", MacGraw-Hill, New York, 1984.
Creve Maples. "Analyzing Software Performance in a Multiprocessor Environment", IEEE Software, 1985, p. 50.
C. MacCann, J. Zahorjan. "Processor Scheduling in Shared Memory Multiprocessors". Technical Report 89,09-17 pp, Depto. of Computer Science and Engineering, University of Washington, September 1989.
Menascé, D. e Almeida, V., "Cost-Performance Analysis of Heterogeneity in Supercomputer Architectures", Proc. ACM-IEEE Supercomputing'90 Conference, New York, November 1990.
V. S. Sunderan. "PVM: A framework for parallel distributed computing. Concurrency: Practice and Experience", 2(4):315-339 pp, December 1990.
