The effectiveness of the Additive Correction Multigrid (ACM) algorithm, a line-byline Tri-diagonal Matrix Algorithm (TDMA), and simple Gauss-Seidel (GS) iteration in numerical heat transfer analysis is investigated on a conventional single processor computer and on a distributed memory parallel computer. The performance of these methods is studied by solving a two-dimensional, steady heat conduction problem. The execution time of ACM on a single processor is proportional to the number of unknowns to the 1.5 power. This is in contrast to the execution time of the TDMA for which the execution time is proportional to the number of unknowns to the 2.0 power. The GS , TDMA and ACM algorithms are adapted to a model IPSC2 Intel hypercube which has a 32 processing nodes each with 8 MBytes oflocal memory. Because GS is a local method, it has almost perfect speed up, but it also converges more slowly than TDMA, The TDMA, on the other hand, is affected by domain decomposition to a greater extent than GS. As the number of processors used to solve the problem is increased, the execution times for GS and TDMA are essentially equal. Solving the model problem with 32 processors on a 192x192 grid resulted in parallel efficiencies of 95%, 80% and 78% for the GS, TDMA, and ACM algorithms, respectively. Though the parallel efficiency of ACM was the lowest of the three, the parallel ACM algorithm required an order of magnitude less time to solve the model than either parallel GS or parallel TDMA without multigrid.
| Identifer | oai:union.ndltd.org:pdx.edu/oai:pdxscholar.library.pdx.edu:open_access_etds-5500 |
| Date | 01 January 1992 |
| Creators | Padgett, James D. |
| Publisher | PDXScholar |
| Source Sets | Portland State University |
| Detected Language | English |
| Type | text |
| Format | application/pdf |
| Source | Dissertations and Theses |
Page generated in 0.002 seconds