A Review of Parallel Methods for Solving Sets of Linear Equations and their Application within Optimisation Algorithms.
Abstract
When solving optimisation problems on a parallel computing system, the first consideration must be to utilize the parallelism to speed up the 95% of the time typically spent in function and gradient evaluations or the division of each function evaluation into a number of parallel tasks. Assuming this prime task has been undertaken effectively then for efficiency the other 5% of the computation must also utilize the parallelism available on the system. The dominant remaining calculation is usually the solution of a set of linear equations. In this paper the implication of parallel processing on the solution method for solving linear equations will be reviewed.
Document Details
- Document Type
- Technical Report
- Publication Date
- Nov 01, 1987
- Accession Number
- ADA193137
Entities
People
- L. C. Dixon