A multithreaded solver for the 2D Poisson equation

Authors:
Andrés Vidal;Alain Kassab;Daniel Mota;Damian Dechev
Affiliations:
University of Central Florida, Orlando;University of Central Florida, Orlando;University of Central Florida, Orlando;University of Central Florida, Orlando
Venue:
Proceedings of the 2012 Symposium on High Performance Computing
Year:
2012

Citing 8
Cited 0

Numerical recipes: the art of scientific computing

Numerical recipes: the art of scientific computing
A parallel solver for unsteady incompressible 3D Navier—Stokes equations

Parallel Computing - Special issue on parallel computing in aerospace
Software Optimization for High Performance Computers

Software Optimization for High Performance Computers
A parallel finite element program on a Beowulf cluster

Advances in Engineering Software - Engineering computational technology
Sourcebook of parallel computing

Sourcebook of parallel computing
MPI implementation of parallel subdomain methods for linear and nonlinear convection--diffusion problems

Journal of Parallel and Distributed Computing
Performance analysis of explicit group parallel algorithms for distributed memory multicomputer

Parallel Computing
Parallel computation of three-dimensional flows using overlapping grids with adaptive mesh refinement

Journal of Computational Physics

Quantified Score

Hi-index	0.00

Visualization

Abstract

A multithreaded solution for the 2D Poisson equation is presented. The proposed algorithm distributes the tasks between threads in Floating-Point Unit (FPU) intensive and non-FPU intensive. This technique also allowed us to make the communication between nodes asynchronous. Our approach of decoupling communication and computation allows for much greater scalability. This new multithreaded approach showed better performance in all multicore processors tested. In the case of the distributed systems tested, the proposed method had greater speed-up than the classical scheme. The technique Red/Black ordering was found to be effective only if data fit entirely in cache memory.