SCIP Solver with multiple threads produces only ABNORMAL results (without any information)

[江林朗]

I want to solve a MIP, but when I set the num of threads, it only produces ABNORMAL results. If I don't set the num of threads, the program can work normally but too slow to get an answer. What can I do if I want to accelerate the program?

[Laurent Perron]

scip is built by the standard cmake command.

On Linux, it is known not to include MT support, while we do specify it on the command line

So it is not working.

Please note that the gain is minimal.

If your problem is purely integral, you can use SAT_INTEGER_PROGRAMMING, which works well with 8 threads or more. 

Le jeu. 28 janv. 2021 à 09:54, 江林朗 <linla...@gmail.com> a écrit :

I want to solve a MIP, but when I set the num of threads, it only produces ABNORMAL results. If I don't set the num of threads, the program can work normally but too slow to get an answer. What can I do if I want to accelerate the program?

--
You received this message because you are subscribed to the Google Groups "or-tools-discuss" group.
To unsubscribe from this group and stop receiving emails from it, send an email to or-tools-discu...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/or-tools-discuss/02ea00e2-4103-4bc7-b8f2-2a8dd4ee5b39n%40googlegroups.com.

[Stuart Rogers]

If you compile SCIP with TPI = tny and if you set the maximum and minimum number of threads parameters parallel/maxnthreads and parallel/minnthreads to n (e.g., 8), then SCIP will run n SCIP solvers in parallel, each in its own thread. You can select the SCIP solver types via the concurrent/scip.../prefprio parameters.

SCIP Doxygen Documentation: How to use the concurrent solving mode (scipopt.org)

https://scipopt.org/doc/html/md_INSTALL.php

https://www.scipopt.org/doc/html/PARAMETERS.php

https://github.com/scipopt/SCIP.jl/issues/242

[Laurent Perron]

We have disabled parallel scip. There are too many errors (leaks, inconsistent results..).

And the gain is minimal. So not worth it.

If your problem is purely integral, use CP-SAT. It is a much better solver, and it benefits greatly from parallelism. 

To view this discussion on the web visit https://groups.google.com/d/msgid/or-tools-discuss/233387e3-eafb-4513-9734-bd0c5c8e3a55n%40googlegroups.com.