Scalable Parallel Algorithms

Abstract

The objective of this project is to develop scaleable parallel formulations of the key computational kernels used in scientific simulations. The specific problems investigated in this project are fast and high quality graph partitioners, highly parallel direct solvers, and parallel formulations of robust preconditioners for iterative solvers, as well as parallel formulations of particle simulation techniques. We have developed a fast and high quality parallel formulations of our multilevel graph partitioning algorithm that are able to partition very large graphs quickly on parallel computers, making it feasible to perform frequent repartitioning of the adaptive and unstructured mesh in adaptive FEM computations. We have developed massively parallel formulations of particle simulation techniques such as Fast Multipole and Barnes-Hut methods, and have investigated the use of this formulation for solving dense linear systems arising in boundary element solution of integral equations. We have developed an MPI-based portable library, called PSPASES, that has been used to solve some of the largest sparse linear systems that have been solved using direct methods. We have also developed robust and parallel preconditioners for iterative solvers using our fast graph partitioning technique and highly parallel Cholesky factorization.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Mar 28, 1999
Accession Number
ADA366733

Entities

People

  • Vipin Kumar

Organizations

  • University of Minnesota

Tags

DTIC Thesaurus Topics

  • Algorithms
  • Boundaries
  • Boundary Element Methods
  • Computational Fluid Dynamics
  • Computations
  • Computer Science
  • Computers
  • Differential Equations
  • Distributed Computing
  • Equations
  • High Performance Computing
  • Information Systems
  • Linear Systems
  • Parallel Computing
  • Parallel Processing
  • Simulations
  • Students

Readers

  • Computational Fluid Dynamics (CFD)
  • Distributed Systems and Data Platform Development
  • Linear Algebra