The open archive for STFC research publications

Full Record Details

Persistent URL http://purl.org/net/epubs/work/63061
Record Status Checked
Record Id 63061
Title Challenges in Parallel Sparse Direct Linear Solvers
Abstract Sparse direct linear solvers are frequently used in both mathematical optimization and finite element methods, often on desktop machines rather than large clusters. While several well-established solvers exist, research avenues in parallel direct solvers are far from exhausted. In particular, challenges remain relating to memory optimizations and the development of practical communication minimization techniques. Of particular interest is achieving strong scaling: good speedup for a fixed size of problem as the number of cores increases. Interior point methods in particular often solve hundreds of linear systems during their execution, and are expected to do so within a few seconds at most. What we see as the main challenges will be summarised, and our current approach to tackling some of them will be presented.
Organisation CSE-NAG , STFC , SCI-COMP
Funding Information
Related Research Object(s):
Licence Information:
Language English (EN)
Type Details URI(s) Local file(s) Year
Presentation Presented at Perspectives on Parallel Numerical Linear Algebra (PPNLA), University of Manchester, UK, 18 Jul 2012. challenges.pdf 2012