Data communication in parallel block predictor-corrector methods for solving ODE's
✍ Scribed by P.S. Rao; G. Mouney
- Publisher
- Elsevier Science
- Year
- 1997
- Tongue
- English
- Weight
- 856 KB
- Volume
- 23
- Category
- Article
- ISSN
- 0167-8191
No coin nor oath required. For personal study only.
✦ Synopsis
In the present article we have considered the problem of data communication in parallel block predictor-corrector (P-BPC) methods for solving ODE's using only time as well as time and space discretizations for systems of equations. After presenting task graphs for each of these discretizations, we present implementation of the time discretized systems of equations on a ring of processors while the time-space discretizations are implemented on a torus network of distributed memory processors. Assuming a store and forward communication model, we have analysed the data communications needed in terms of the number of steps and the bandwidth needed for a single cycle of the P-BPC algorithm. The expressions are presented for fill-duplex communication links. For torus implementation, we propose several alternate methods, with the intention of minimizing the total communication per cycle of the P-BPC algorithm. 0 1997 Published by Elsevier Science B.V.