© 2002 Systems Research Group, Department of Computer Science and Information Systems, The University of Hong Kong.

Main ** People ** Research ** Publications ** Facilities ** Links

High Performance Irregularly Structured Problems

The irregularly structured problems (ISP), where the patterns of computation and communication are unstructured and/or changing dynamically, arises in parallelizing many large-scale applications. These include unstructured finite element methods, molecular dynamic, N-body simulations, multimedia communication, etc. Even though the current generation of supercomputers, such as Cray T3D, IBM SP-2, Intel Paragon, and ATM-based clusters of workstations/servers, can achieve hundreds Gigaflops to Teraflops of raw computing power. The lack of efficient algorithmic techniques in using these machines makes the large speed-up unattainable. The objective of this project is to develop scalable algorithms for these problems as well as portable implementations of the selected applications on commodity workstation or PC clusters. We aims at achieving interactive latencies for applications where the order of a few seconds are permissible, comparing to current serial methods which provide latencies of several minutes.

Back to Research Main page.

Technical Papers:

• Y. Sun and C.L. Wang, "A Distributed Object Model for Solving Irregularly Structured Problems on Clusters", International Conference on Cluster Computing, 8-11 Oct. 2001, Newport Beach, pp.187-190.
• Y. Sun and C.L. Wang. "A Multithreaded Framework for Distributed Computing on Non-Uniform SMP Cluster", submitted to 29th International Conference on Parallel Processing, (ICPP 2000), Toronto, Canada, August 21-24, 2000.
• Y. Chung, C.L. Wang, and V.K. Prasanna. "Parallel Algorithms for Perceptual Grouping on Distributed Memory Machines", Journal of Parallel and Distributed Computing, pp. 123-143, Vol. 50, No. 1/2, April 10/May 1, 1998.

Main ** People ** Research ** Publications ** Facilities ** Links