Computing Seminar

"Design Strategies for
Irregularly Structured Parallel Applications"

Leonid Oliker, LBNL

Abstract

The success of parallel computing in solving real-life computationally intensive problems relies on their efficient mapping and execution on large-scale multiprocessor architectures. Many important applications are both unstructured and dynamic in nature, making their efficient parallel implementation a daunting task.

In this talk I will present the parallelization of two irregularly structured computations: the Conjugate Gradient algorithm and unstructured mesh adaptation. Both codes were developed with an MPI message passing implementation on the Cray T3E and the SGI Origin2000, a shared-memory implementation using the cache coherent nonuniform memory access (CC-NUMA) of the Origin2000, and a multithreaded version on the newly released Tera Multithreaded Architecture (MTA). A comparison is presented between the critical factors of these parallel code developments, including runtime, scalability, programmability, and memory overhead. The overall results demonstrate that multithreaded systems offer a tremendous potential for quickly and efficiently solving some of the most challenging real-life problems on parallel computers.

This work won the Best Paper Award at Supercomputing99.

About the speaker: Leonid Oliker is currently a computer scientist in the Future Technologies Group at the National Energy Research Scientific Computing Center (NERSC), located at Lawrence Berkeley National Laboratory (LBNL). In 1991, he received a B.S.E. in computer engineering from the University of Pennsylvania and a B.S. in finance from the Wharton School of Business. He received his Ph.D. in computer science from the University of Colorado in 1998. Dr. Oliker has held positions as a visiting researcher and a postdoctoral scientist at the Research Institute for Advanced Computer Science (RIACS) at NASA Ames Research Center. His research interests include the study of dynamically adapting algorithms on advanced parallel architectures, job scheduling for effective system performance, generic programming for scientific computations, processor-in-memory clusters, and resource management for mobile computing.