SC 740 Review

By

Deborah Dent

Department of Computer Science

Mississippi State University

and

NSF ERC for Computational Field Simulation

Mississippi State, Mississippi

Friday, May 8th, 1998

Dr. Banicescu presented a talk on simulating the evolution of N particles over time given initial position and velocities. After stating the problem, Dr. Banicescu pointed out several applications such as astrophysics and molecular biology. After listing several approximation algorithms, she presented details on the Greengard algorithm.

Dr. Banicescu next presented details on parallelization. The subtrees that are used in the simulations are distributed among several processors. Performance issues included:

• Accuracy
• Speedup
• Scalability
• Simplicity
• Robustness

Other issues include mapping, load balancing and communications. It appears that performance gains are difficult to obtain due to load imbalances caused by the irregular distribution of bodies. She then defined fractiling as a dynamic scheduling scheme that simultaneously balances processor loads and maintains locality by exploiting the self-similarity properties of fractals. After presenting details on fractals, she reported on experiments, which were performed, on the IBM_SP2 and the SuperMSPARC. The code used MPI and was based on the PFMA code from Duke University.

After presenting implementation highlights and experimental results, we were informed that performance improvements were obtained on uniform and nonuniform distribution of bodies, underscoring the need for a scheduling scheme that accommodates application as well as system induced execution time variance.

In conclusion, Dr. Banicescu stated that this work has effective techniques for Scientific Computing applications. In fact, she has received requests from IBM and Bell Labs for the future work.