|Subject:||Re: [Help-glpk] Experience with GLPK in OpenMP|
|Date:||Fri, 19 Aug 2011 09:21:35 -0700|
> Does anyone have any experience in compiling and running glpk with
> OpenMP and on a multi-cored computer, such as an Intel i7.
> What are the speed factors achieved for the optimization problems?
What are you thinking of parallelizing? The various simplex flavors are not inherently parallel algorithms, so there isn't a simple way to 'turn-on' parallelization.
If you are thinking about branch-and-bound stuff for integer optimization, then I will have to defer to others on the list, but I don't think glpk is parallelized for that either.
However, if you have something more specific in mind, let us know. As a shameless plug, if you are considering something like Dantzig-Wolfe decomposition, I have a parallel implementation of that on sourceforge (search for dwsolver) built on top of GLPK. For the right problems, you can get massive speedups (100's of times faster) and can actually solve faster than a base CPLEX implementation. It uses pthreads rather than OpenMP.
There are issues with GLPK and thread-safety/re-entrant behavior which are pretty well documented on this help list that you need to consider before going for a parallel implementation.
|[Prev in Thread]||Current Thread||[Next in Thread]|