> Can I run LP solver in parallel on multiple cores? What about all running on GPUs?
All major LP solvers include a barrier/interior point code that can run on multiple cores. This includes Cplex, Gurobi and Mosek. Do not expect perfect scaling but some speed-up is possible. Sparse simplex algorithms are not ideally suited for parallelization. Some recent results were shown by John Forrest. See: ISMP 2012 talk.
About GPUs:
(From the same talk).
There are some other algorithms that can be implemented on GPUs. Many meta-heuristics have been ported to these devices. Here are some links:
- http://www.fit.vutbr.cz/research/view_pub.php?file=%2Fpub%2F9184%2F60240442.pdf&id=9184
- http://link.springer.com/chapter/10.1007%2F978-3-642-04843-2_4?LI=true
- http://www.gputechconf.com/content/GTC/posters/2010/A04-Parallel-Ant-Colony-Optimization-with-CUDA.pdf
- http://sikaman.dyndns.org:8888/honours/karim-tantawy-winter-2011.pdf
- and many more
No comments:
Post a Comment