Compiler-optimized simulation of large-scale applications on high performance architectures

  • Authors:

  • Vikram S. Adve;Rajive Bagrodia;Ewa Deelman;Rizos Sakellariou

  • Affiliations:

  • University of Illinois at Urbana-Champaign, Urbana, Illinois;University of California at Los Angeles, Los Angeles, California;Information Sciences Institute, University of Southern California, Marina del Rey, California;University of Manchester, Manchester M13 9PL, United Kingdom

  • Venue:
  • Journal of Parallel and Distributed Computing - Parallel and Distributed Discrete Event Simulation--An Emerging Technology
  • Year:
  • 2002

Quantified Score

Hi-index 0.00

Visualization

Abstract

In this paper, we propose and evaluate practical, automatic techniques that exploit compiler analysis to facilitate simulation of very large message-passing systems. We use compiler techniques and a compiler-synthesized static task graph model to identify the subset of the computations whose values have no significant effect on the performance of the program, and to generate symbolic estimates of the execution times of these computations. For programs with regular computation and communication patterns, this information allows us to avoid executing or simulating large portions of the computational code during the simulation. It also allows us to avoid performing some of the message data transfers, while still simulating the message performance in detail. We have used these techniques to integrate the MPI-Sim parallel simulator at UCLA with the Rice dHPF compiler infrastructure. We evaluate the accuracy and benefits of these techniques for three standard message-passing benchmarks on a wide range of problem and system sizes. The optimized simulator has errors of less than 16% compared with direct program measurement in all the cases we studied, and typically much smaller errors. Furthermore, it requires factors of 5 to 2000 less memory and up to a factor of 10 less time to execute than the original simulator. These dramatic savings allow us to simulate regular message-passing programs on systems and problem sizes 10 to 100 times larger than is possible with the original simulator, or other current state-of-the-art simulators.