A structural analysis algorithm for massively parallel computers
We describe a high performance parallel algorithm based on the conjugate gradient method for large structural analysis problems. The algorithm has been tested on a 1024-processor hypercube. Three performance models are measured: 1024-processor speedup is 502 when problem size is fixed, 987 when execution time is fixed, and 1019 when problem size per processor is fixed. The latter two models best show the practical capabilities of large ensembles. Sustained performance is 132 MFLOPS (64-bit), matching or surpassing performance of more conventional processing methods. Measurements show the techniques will extend to even higher levels of parallelism than the 1024-processor level explored here. 11 refs., 3 figs., 4 tabs.
- Research Organization:
- Sandia National Labs., Albuquerque, NM (USA)
- DOE Contract Number:
- AC04-76DP00789
- OSTI ID:
- 6525354
- Report Number(s):
- SAND-88-3363C; CONF-8810280-2; ON: DE89005055
- Country of Publication:
- United States
- Language:
- English
Similar Records
Massively parallel mathematical sieves
A scalable parallel algorithm for sparse Cholesky factorization