Title: Task assignment heuristics for parallel and distributed CFD applications

Authors: Noe Lopez-Benitez, M. Jahed Djomehri, Rupak Biswas

Addresses: Texas Tech University, Lubbock, 79409-3104 TX, USA. ' Computer Sciences Corporation, NASA Ames Research Center, Moffett Field, 94035-1000 CA, USA. ' NASA Advanced Supercomputing Division, NASA Ames Research Center, Moffett Field, 94035-1000 CA, USA

Abstract: A Task Graph (TG) model is proposed for representing a single discrete step of multi-block overset grid Computational Fluid Dynamics (CFD) applications. The TG model is used to balance the computational workload across the overset grids and to reduce inter-grid communication costs. Based on the constraints inherent in CFD applications, assignment heuristics are developed and enhanced by integrating the status of processing units and communication costs. Extensive performance evaluation on a synthetic TG is reported. Also, a TG derived from a realistic problem with eight million grid points is used as a test case.

Keywords: overset grids; task graphs; TG; performance prediction; parallel processing; high performance computing; parallel CFD; distributed CFD; computational fluid dynamics; task assignment.

DOI: 10.1504/IJCSE.2007.015745

International Journal of Computational Science and Engineering, 2007 Vol.3 No.2, pp.155 - 165

Published online: 11 Nov 2007 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article