Usando Performance Relativa para Avaliar Desempenho em Supercomputadores Paralelos
Resumo
Em principio, a performance de um supercomputador paralelo é o resultado da atuação do escalonador sobre a workload que está sendo processada. Entretanto, dependendo da métrica utilizada para representar este desempenho, o resultado final da avaliação pode variar, gerando conflitos. Dessa forma, a métrica de performance é também um fator que influencia nos resultados do processo de avaliação. A métrica Performance Relativa, aqui apresentada, através da comparação dos turn-around times de todo o conjunto de programas obtidos pelos diferentes métodos de escalonamento, possibilita minimizar a influência que a métrica possa exercer sobre o resultado da avaliação de performance. Possibilita também identificar e estudar os subconjuntos de dados onde cada método obteve melhor desempenho, a fim de determinar suas características.
Referências
D. Feitelson, The Effect of Metrics and Workloads on Performance Evaluation. Technical Report 2001-13, School of Computer Science and Engineering, The Hebrew University of Jerusalem. Oct 2001.
V. Lo, J. Mache, and K. Windisch. A Comparative Study of Real Workload Traces and Synthetic Workload Models for Parallel for Scheduling. In Job Scheduling Strategies for Parallel Processing, Dror Feitelson and Larry Rudolph (Eds.), Springer Verlag, Lect. Notes Comput. Sci. vol. 1459, pp. 25-46,1998.
D. Feitelson. Analyzing the Root Causes of Performance Evaluation Results. Technical Report 2002-4, School of Computer Science and Engineering, The Hebrew University of Jerusalem, Mar 2002.
S. Chapin, W. Cirne, D. Feitelson, J. Jones, S. Leutenegger, U. Schwiegelshohn, W. Smith, and D. Talby. Benchmarks and Standards for the Evaluation of Parallel for Schedulers. In Job Scheduling Strategies for Parallel Processing, D. Feitelson and Larry Rudolph (Eds.) Springer-Verlag, Lecture Notes in Computer Science, vol. 1659, pp. 66-89, 1999.
D. Feitelson and A. Mu'Alem. Utilization and Predictability in Schedulling IBM SP2 with Backfilling. In 12th Int. Parallel Processing Symp., pp 542-546, Apr 1998.
A. Mu'Alem and D. Feitelson. Utilization, Predictability, Workloads, and User Runtime Estimates in Scheduling the IBM SP2 with Backfilling. IEEE Trans. Parallel & Distributed Syst, 12(6), pp. 529-543, Jun 2001.
W. Cirne, Using Moldability to Improve the Performance of Supercomputer Jobs. Ph.D. Thesis, University of California, San Diego. 2001.
P. L. Costa Neto. Estatística. Editora Edgard Blücher Ltda, l977.
D. Feitelson, Metrics for Parallel Job Scheduling and Their Convergence. In Job Scheduling Strategies for Parallel Processing, D. G. Feitelson and L. Rudolph (Eds.), SpringerVerlag, Lecture Notes in Computer Science Vol. 2221, pp. 188- 206, 2001.
Standard Performance Evaluation Corporation. The SPEC web page. http://www.spec.org/
Dror Feitelson. The Parallel Workloads Archive. http://www.cs.huji.ac.il/labs/parallel/workload/
Victor Hazlewood. NPACI ]obLog Repository. http://joblog.npaci.edu/
W. Cirne and F. Berman. A Comprellensive Model of the Supercomputer Workload. In Proceedings of WWC-4: IEEE 4th Annual Workshop on Workload Characterization, Dec 2001.
