We ran out of memory trying to solve your example with the setting threads=32. However we were successful with threads=1 and threads=2 as you can see in the attached listing. You can see that threads=2 did not speed up the solution very much. Most of the computation was in root node processing, which is not readily parallelizable; the tree search, which is the part of the algorithm that lends itself to distribution over multiple threads, was easy for threads=1 and not even needed for threads=2.
Each extra thread does create a requirement for more memory, so in general it is a good idea to try threads=1, 2, 4, 8, ... rather than trying to use a very large number of threads from the beginning. Even for problems that require CPLEX to spend a substantial amount of time in the tree search, there tends to be limited benefit to adding threads past 8 or 16 -- see for example the mixed-integer linear programming benchmarks at
http://plato.asu.edu/bench.html.
Bob Fourer
am...@googlegroups.com
=======