Originally posted by: BenjaminKlotz
Hi,
I am using the CPLEX c interface, loading CPLEX as a dll.
I am solving many MIP instances concurrently, starting one CPLEX instance per thread.
The standard malloc on my system performs poorly. As a result, my approach does not scale past 30 threads, due to contention in malloc performed by CPLEX.
Is there a way to configure CPLEX to use a custom allocator?
I am aware of the option to preload a libc replacement. Unfortunately, this is not feasible for me, as my code (and also CPLEX) is loaded as a dll as part of a bigger flow that I don't control.
Any suggestions are appreciated :)
Ben
#CPLEXOptimizers#DecisionOptimization