Sensitivity analysis of parallel applications to local and non-local interference
The environment in which a parallel application is executed has high impact on the performance of the application due to interference caused by various factors in the execution environment. A detailed understanding of the sensitivity of the application to the parameters describing the execution environment can be of great help in (a) predicting a suitable target machine model for the application, (b) predicting its performance on the target machine, and (c) any algorithmic bottlenecks. In this paper, we analyze a suite of parallel applications for their sensitivity to local and non-local interference arising due to various factors in a parallel environment. We create a test bed consisting of five different parallel applications taken from different sources and analyze their sensitivity to single node and multi-node perturbations and show that parallel applications can behave very differently under different conditions of interference in the environment in which they are running. The main contributions of this paper are: (a) studying a suite of parallel algorithms for their sensitivity to local and non-local interference, (b) demonstrate that an application can behave differently to different interference levels in the environment, (c) demonstrate that the sensitivity of an application can be quantified as its absorption ratio at a given interference level.
KeywordsSingle Node Message Passing Interface Parallel Application Execution Environment Interference Level
Unable to display preview. Download preview PDF.
- D. Bader. A practical parallel algorithm for cycle detection in partitioned digraphs. Technical Report UNM/AHPCC99- 013, 1999.Google Scholar
- D. Bader and J. Ja’Ja’. Practical parallel algorithms for dynamic data redistribution, median finding, and selection. 10th International Parallel Processing Symposium (IPPS 96) Conference, Honolulu, HI, April 15-19, pages 292.301, 1996.Google Scholar
- R. M. Badia, J. Labarta, J. Gim’enez, and F. Escal’e. DIMEMAS: Predicting MPI applications behavior in Grid environments. Workshop on Grid Applications and Programming Tools, 8 th Global Grid Forum (GGF8), Seattle, WA, 2003.Google Scholar
- Message Passing Interface Forum. MPI: A message-passing interface standard. Technical Report UT-CS-94-230, University of Tennessee, 1994.Google Scholar
- R. Mraz. Reducing the variance of point to point transfers in the IBM 9076 parallel computer. In Proceedings of the 1994 conference on Supercomputing, 1994.Google Scholar
- J. Nieplocha and B. Carpenter. Armci: A portable remote memory copy library for distributed array libraries and compiler run-time systems. Lecture Notes in Computer Science, 1999.Google Scholar
- U. of California. http://setiathome.berkeley.edu/.Google Scholar
- F. Petrini, D. Kerbyson, and S. Pakin. The Case of the Missing Supercomputer Performance: Achieving Optimal Performance on the 8,192 Processors of ASCI Q. In Proceedings of Supercomputing 2003, 2003.Google Scholar
- M. Sottile and R. Minnich. Analysis of Microbenchmarks for the Performance Tuning of Clusters. In Proceedings of Cluster 2004, 2004.Google Scholar
- M. J. Sottile, V. Chandu, and D. A. Bader. Performance analysis of parallel programs via message-passing graph traversal. 20th IEEE Int’l Parallel and Distributed Processing Symp. (IPDPS), Rhodes Island, Greece, April 25-29, 2006.Google Scholar
- I. web page. www.intel.com/software/products/cluster/clustertoolkit/index.htm.Google Scholar