I am working with a large scale MIP which could be solved to optimal by CPLEX, however it takes a lot of time computationally (~8 hours for 41 trucks starting from 346 depots delivering 197 customers) and would like to discuss how to improve the efficiency here. I want to know,
Please share your experiences on what kind of parameter tuning helped to accelerate the solution process. Also any paper reference on reformulating the constrains to obtain quicker solutions would be helpful. asked 12 Aug '14, 16:22 Pavan
showing 5 of 7
show 2 more comments

Sounds like you are solving a reasonably sized vehicle routing problem. You should be able to get better runtime than what you are seeing right now. It does not look like your problem is extremely big or dense. It could be that your numerics are bad, maybe due to the bigMs or due to small values you have somewhere in your matrix? To your questions:
This paper might also be helpful in finding a good formulation. answered 13 Aug '14, 13:25 Philipp Chri... 
BigM formulations quite often lead to very weak linear relaxations. Especially if you use one single M, like M=1mio, without considering better (smaller) values of M. If you have more than one of these bigM constraints, you might want to try to index them, the bigM's, such that they are tailored to the specific constraint in question. I came by this paper the other day: http://www.optimizationonline.org/DB_HTML/2014/08/4483.html. I haven't read it, but it might be useful.
The link accidentally included the period ending the sentence. Trim that off and the link works.
yeah got it now
Regarding 1., looking into the answers to https://www.orexchange.org/questions/19/goodwaystoworkaroundthebigintegralitygapcausedbybigmformulationsinmipproplems might pay off.
Do you mean "8 hours for 197" NODES or does your problem really only have 197 variables. If not, how many nodes does CPLEX process and how many constraints does your problem it have?
This thread is going no where without a log dump :)
@Philipp Christophel: After presolve, 18840 rows, 1082140 columns, and 2635268 nonzeros.