Performance prediction of large-scale parallel discrete event models of physical systems

ABSTRACT

A virtualization system is presented that is designed to help predict the performance of parallel/distributed discrete event simulations on massively parallel (supercomputing) platforms. It is intended to be useful in experimenting with and understanding the effects of execution parameters, such as different load balancing schemes and mixtures of model fidelity. A case study of the virtualization system is presented in the context of plasma physics simulations, highlighting important virtualization challenges and issues, such as reentrancy and synchronization in the virtual plane, and our corresponding solution approaches. A trace-based prediction methodology is presented, and is evaluated with a 1-D hybrid collisionless shock model simulation, with the predicted performance being validated against one obtained in actual simulation. Predicted performance measurements show excellent agreement with actual performance measurements on parallel platforms containing up to 512 CPUs.

REFERENCES

• 1: Colella, P., D. T. Graves, T. J. Ligocki, D. F. Martin, D. Modiano, D. B. Serafini, and B. V. Straalen. 2005. Chombo software package for AMR applications: Design Document. http://seesar.lbl.gov/anag/chombo/.
• 2: Phillip M. Dickens , Philip Heidelberger , David M. Nicol, Parallelized Direct Execution Simulation of Message-Passing Parallel Programs, IEEE Transactions on Parallel and Distributed Systems, v.7 n.10, p.1090-1105, October 1996 [doi>10.1109/71.539740]
• 3: H. Karimabadi , J. Driscoll , Y. A. Omelchenko , N. Omidi, A new asynchronous methodology for modeling of physical systems: breaking the curse of courant condition, Journal of Computational Physics, v.205 n.2, p.755-775, 20 May 2005 [doi>10.1016/j.jcp.2004.12.003]
• 4: Karimabadi, H., J. Driscoll, Y. Omelchenko, K. S. Perumalla, R. M. Fujimoto, and N. Omidi. 2005. Parallel discrete event simulation of grid-based models: asynchronous electromagnetic hybrid code, Lecture Notes in Computer Science, in press.
• 5: King, S. T., G. W. Dunlap, and P. M. Chen. 2003. Operating system support for virtual machines, presented at Annual USENIX Technical Conference.
• 6: Jason Liu , David Nicol , Brian J. Premore , Anna L. Poplawski, Performance prediction of a parallel simulator, Proceedings of the thirteenth workshop on Parallel and distributed simulation, p.156-164, May 01-04, 1999, Atlanta, Georgia, United States
• 7: Friedemann Mattern, Efficient algorithms for distributed snapshots and global virtual time approximation, Journal of Parallel and Distributed Computing, v.18 n.4, p.423-434, Aug. 1993 [doi>10.1006/jpdc.1993.1075]
• 8: Omelchenko, Y. 2005. Scientific discrete event simulation (SciDES) tools, SciberQuest Inc., Technical Report.
• 9: Kalyan S. Perumalla, µsik " A Micro-Kernel for Parallel/Distributed Simulation Systems, Proceedings of the 19th Workshop on Principles of Advanced and Distributed Simulation, p.59-68, June 01-03, 2005 [doi>10.1109/PADS.2005.1]
• 10: Sundeep Prakash , Rajive L. Bagrodia, MPI-SIM: using parallel simulation to evaluate MPI programs, Proceedings of the 30th conference on Winter simulation, p.467-474, December 13-16, 1998, Washington, D.C., United States
• 11: Steven K. Reinhardt , Mark D. Hill , James R. Larus , Alvin R. Lebeck , James C. Lewis , David A. Wood, The Wisconsin Wind Tunnel: virtual prototyping of parallel computers, Proceedings of the 1993 ACM SIGMETRICS conference on Measurement and modeling of computer systems, p.48-60, May 10-14, 1993, Santa Clara, California, United States
• 12: Terry L. Wilmarth , Gengbin Zheng , Eric J. Bohm , Yogesh Mehta , Nilesh Choudhury , Praveen Jagadishprasad , Laxmikant V. Kale, Performance Prediction Using Simulation of Large-Scale Interconnection Networks in POSE, Proceedings of the 19th Workshop on Principles of Advanced and Distributed Simulation, p.109-118, June 01-03, 2005 [doi>10.1109/PADS.2005.20]
• 13: Zheng, G., G. Kakulapati, and L. V. Kale. 2004. BigSim: a parallel simulator for performance prediction of extremely large parallel machines, IPDPS.