I am solving a large-scale feasibility optimization problem on a computational node with multiple cores. My problem (seemingly) is NLP without mixed integer constraints. When I solve it using Baron, it takes a long time. In the Baron documentation:
the only choice related to parallelization is “threads” , but it seems only to be related to MIP. I tried changing the default which is 1 to say 50 but that did not make any difference.
My question is How to make Baron solves the problem more efficiently by taking advantage of me solving in a computational node?
BARON uses the parallel capabilities of the integrated MIP solvers (e.g. Cplex) when it has to solve MIP relaxations (for MINLPs). For pure continuous problems BARON uses LP relaxations and can’t benefit from the “threads>1” option. BARON itself seems not to have it’s own parallel capabilities (yet).
Thanks. I just checked my workstation. The only global solver available to me is ANTIGONE. I was also advised about a choice in the Cplex option file called “LP method” and I can set it to be “concurrent optimization”. I looked that up but couldn’t find anything related
1- Looking at ANTIGONE documentation, I found the option to pass a cplex option file named “cplex_optfile”. My question is how to write such an option file? what is the format? do I have to include all options in that file? or only the options that I want to change?
2- Does changing the LP algorithm to, say, Barrier, rather than Simplex help making the solving faster?
It seems that even Baron uses cplex as a subsolver. However, I couldn’t tell based on Baron documentation if you can pass cplex option file. Can we do that?