Originally posted by: SystemAdmin

[factorial said:]

I'm trying to solve a big MIP problem using Lagragian relaxtion, and I use the subgradient method approach. So, I have many subgradient iterations times many (decomposed) subproblems to solve.

What I did is "write an LP" -> "open Cplex env - solve - close Cplex env" -> "write another LP" -> "open Cplex env - solve - close Cplex env" -> .....
Will this be memory intensive?
Shoud I open the Cplex env only once and solve for all lp's and then finally close the Cplex env?

Because in some cases, I got the error message "Could not open CPLEX environment." with CPLEX Error 32201: ILM Error 7: CPLEX: no access key found for this product and CPLEX Error 32201: ILM Error 16: CPLEX: license file not found or unreadable.
I'm sure that license is not an issue but I'm wondering if my implementation causes this problem.

Hope someone can help! Thanks.

#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

[prubin said:]

It's been a while since I had to deal with environments -- they're handled implicitly, not explicitly, in the Java API -- but IIRC opening multiple environments counts as multiple uses of the CPLEX license, so if you are licensed for only one user at a time, that's probably the source of the licensing error message.  (There is probably some point at which more than one environment is open simultaneously.)  In any case, I would generally use a single environment unless I had a compelling reason to the contrary.

I would also leave the LPs alive from iteration and add/remove constraints or variables as needed, rather than recreating the entire LP each time.  Making incremental changes is usually faster than rebuilding every LP from scratch.

/Paul

#CPLEXOptimizers
#DecisionOptimization