DISTRIBUTION-INDEPENDENT FRAMEWORK FOR SCIENTIFIC APPLICATIONS

ABSTRACT:

In many scientific applications, the behavior of a physical system is captured through either 1) a discretization of the space using a finite grid, followed by determining the quantities of interest at each of the grid points/cells or 2) a simulation of the motion of (real or virtual) particles whose evolution can be related to the behavior of the system. The fundamental challenge posed by these applications is that the run-times of algorithms depend not only on the number of unknowns but also on their distribution. In this talk, I will describe a distribution-independent framework we have designed for efficient solution of such applications on sequential and parallel computers. I will illustrate the framework in the context of the N-body problem, considered to be a "grand challenge" problem. I will present the first provably efficient and distribution-independent parallel algorithm for the N-body problem. Our work has the side effect of showing that the N-body problem is not irregular, as has been thought for more than a decade.

  Prof. Srinivas Aluru is a faculty member in the Department of Electrical and Computer Engineering at Iowa State University. He is interested in application-driven research in sequential and parallel computing, scientific computing and computational biology. Prior to joining Iowa State, he held a faculty position at New Mexico State University for 3 years and a visiting faculty position at Syracuse University for 2 years. He received his B.Tech degree in Computer Science and Engineering from Indian Institute of Technology, Madras, in 1989 and his M.S. and Ph.D. degrees in Computer Science from Iowa State University in 1991 and 1994. He is a recipient of the NSF CAREER award.