Performance improvement of asynchronous iterations by non-uniform load distribution
Conference
·
OSTI ID:125566
- Pacific Northwest Lab., Richland, WA (United States)
- Univ. of Alabama, Tuscaloosa, AL (United States)
Asynchronous iterative methods may be employed to solve large and sparse systems of linear equations that arise in the numerical solution of partial differential equations by using finite difference methods. Asynchronous iterations have been studied in hope that they would offer better ways for dealing with the high communication latency on the massively parallel systems. Comparing to their synchronous counterparts, asynchronous iterations may be indeed implemented more efficiently due to the lack of synchronization; however, their convergence rate could be worse than expected. This paper presents a nontraditional approach to load balancing as a way to improve performance of asynchronous iterations.
- OSTI ID:
- 125566
- Report Number(s):
- CONF-950212--; CNN: Grant 5-31433
- Country of Publication:
- United States
- Language:
- English
Similar Records
Asynchronous Richardson iterations: theory and practice
Modeling the asynchronous Jacobi method without communication delays
Massive Asynchronous Parallelization of Sparse Matrix Factorizations
Journal Article
·
Sat Oct 17 20:00:00 EDT 2020
· Numerical Algorithms
·
OSTI ID:1832741
Modeling the asynchronous Jacobi method without communication delays
Journal Article
·
Thu Feb 21 19:00:00 EST 2019
· Journal of Parallel and Distributed Computing
·
OSTI ID:1527000
Massive Asynchronous Parallelization of Sparse Matrix Factorizations
Technical Report
·
Sun Jan 07 23:00:00 EST 2018
·
OSTI ID:1416151