A Framework for Matching Applications with Parallel Machines

Investigators: S. Sahni, S. Ranka, and J. Peir

Abstract:

This project is to assess performance of a selected set of ITL/CEWES applications on various supercomputers. Instead of measuring the run time of the applications on each machine (using different number of processors), we estimate the execution time based on characterization of the workload. This includes two parts. The first is to develop basic computational structures that can be extrapolated according to the number of processors and the problem size. The second part is to model precisely the communication and synchronization requirement. Performance measurement on the target machines can be used for validation of the modeling and prediction results.

Papers and Presentations:

  1. J. In, C. Jin, J. Peir, S. Ranka, and S. Sahni, "A framework for matching parallel applications with parallel machines," 6th International Conference on High Performance Computing, Lecture Notes in Computer Science, Springer Verlag, Volume 1745, Dec. 1999