Efficient Load Balance & Locality of Reference for Unstructured Grid & Particle Simulations on Massively Parallel Processors.
Abstract
During the contract period our main results are a computer code for fast parallel algorithms for particle systems interacting with long range forces, analysis of the error characteristics of the chosen method, and a parallel implementation of a O(N log(2 over 2) N) algorithm for Legendre and Spherical transforms. We have also derived an algebraic framework for describing permutations frequently used in scientific computation. The framework allows for a rigorous analysis of the communication requirements of parallel algorithms and is also very useful in address computations during compilation or in run time systems. For efficient data motion, or remote references, we have also further validated the potential benefits of ROMM routing.
Document Details
- Document Type
- Technical Report
- Publication Date
- Nov 30, 1995
- Accession Number
- ADA311016
Entities
People
- S. L. Johnsson
Organizations
- Harvard University