Thursday, September 21

Load Balancing for Parallel Applications - Part I of II

10:30 AM-12:30 PM
Center City 1

For Part 11, see MS16.

Load balancing is vital in parallel computation. To maximize parallel efficiency, problems must be divided among processors so that work is evenly distributed and communication costs are low. For applications such as finite element and finite difference methods, static partitioners are often used to obtain initial decompositions. Dynamic load balancing is needed when applications' work loads or communication patterns change during computations, as in adaptive finite element methods, particle methods, and contact detection algorithms. In this minisymposium, the speakers will discuss recent development of load-balancing tools and their use in parallel computational applications; both users and developers of load balancing tools should benefit.

Organizers: Karen D. Devine and Erik G. Boman
Sandia National Laboratories, Albuquerque, USA
10:30-10:55 Zoltan: A Dynamic Load Balancing Library for Parallel Computational Applications
Erik G. Boman, Karen D. Devine, Bruce Hendrickson, and Courtenay Vaughn, Sandia National Laboratories, Albuquerque, USA
11:00-11:25 New Formulations for Adaptive Graph Partitioning
Kirk Schloegel, George Karypis, and Vipin Kumar, Army HPC Research Center, University of Minnesota, Minneapolis, USA
11:30-11:55 Integrated Load Balancing and Data Management for Adaptive Finite Element Schemes
Abani Patra, Jingping Long, and Andrew C. Bauer, State University of New York at Buffalo, USA
12:00-12:25 Performance of Partitioning Metrics and Comments on Their Applications for Adaptive Algorithms
Graham F. Carey, Atanas I. Pehlivanov, and Saeed Iqbal, University of Texas at Austin, USA

©2000, Society for Industrial and Applied Mathematics
Designed by Donaghy's Web Consulting
Created 6/14/00; Updated 6/14/00