Originally posted by: SystemAdmin

[ph mooraa said:]

Thanks to the advice of group members, I was able to make my CPLEX code work.

But it runs for only small size problem. for larger sets it gives error "not enough memory" on 2GB red hat machine. I have few questions about it.

(1) how to know whether my problem is solvable in terms of memory requirement? the decision variables are 350x350x1500000 sized binary array. another array of 1500000 too. there are 130000 constraints plus few more approximately. Is such a problem at all possible to solve using CPLEX? if yes, what would be the memory requirement? any experienced member might be knowing this, in case.

related answer in FAQ is sort of unclear to me since it is in terms of constraints only.

(2) after looking into manual and searching on web, I tried adding following to the code

                mycplex.setParam(IloCplex::MemoryEmphasis,1);
                mycplex.setParam(IloCplex::WorkDir,".");
                mycplex.setParam(IloCplex::WorkMem,1000000);
                mycplex.setParam(IloCplex::NodeFileInd,2);

but these seem to make no difference at all. Is there something else required to add which can enable inbuilt CPLEX options to better utilize memory. I donot have limitation over time. is there a way to make CPLEX run slower and thereby may be occupy lesser memory?

(3) what is difference between "ERROR:not enough memory" and "out of memory" errors? the latter seems to be coming after a while if the problem size is smaller while former comes immediately after a huge problem (like mentioned above) is started.

any advice is appreciated.
Thanks, phm
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

[jgregory said:]

If I interpret correctly, when you say "the decision variables are 350x350x1500000 sized binary array", you want your model to make around 183 billion binary decisions.  That is several orders of magnitude larger than one can reasonably expect to solve, limited first by computer memory size but also by the ability of the algorithms.  A model with "only" a million binary variables (in a single dimension) to go with the 130000 constraints you mention would still be considered large and very difficult unless good luck is with you in the combinatorial aspects.
#CPLEXOptimizers
#DecisionOptimization