Originally posted by: SystemAdmin

[aroso said:]

Hi,

I have a program running CPLEX several times in a row. My problem is that after the first couple of runs, CPLEX starts taking a very long time to start up (around 30 minutes) after the "Reduced MIP has ..." information is displayed.

However, the presolve and root relaxation time displayed seem normal:
Tried aggregator 2 times.
MIP Presolve eliminated 530 rows and 411 columns.
MIP Presolve modified 10 coefficients.
Aggregator did 10 substitutions.
Reduced MIP has 2701 rows, 2210 columns, and 19100 nonzeros.
Presolve time =    0.13 sec.
Clique table members: 1201.
MIP emphasis: balance optimality and feasibility.
Root relaxation solution time =    0.23 sec.


Shouldn't those value be higher? And why is it taking so long after a couple of runs?

Thanks in advance,
Antonio




#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

[arunjotshi said:]

Did you try emphasizing feasibility rather than optimality.
Sometimes, depending on the problem structure, CPLEX will be trying to look more and more for optimal solutions rather than good feasible solutions.
You might also want to favor branching on either 1 or 0, depending on your problem structure.
Hope this helps.
Thanks,
Arun
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

[aroso said:]

Thank you for your answer.

However, my problem is not in finding a solution - it's in the time cplex takes to start the branch and bound process.

Antonio
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

[Didier Vidal said:]

Antonio,

Maybe this page could give you useful information then. It's about ILOG CPLEX Performance Tuning for Mixed Integer Programs (MIPs)

http://www.ilog.com/products/optimization/info/mipperformance.cfm

Didier.
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

[MaryFenelon said:]

You can get more information about what is happening during the root solve by setting the MIPDisplay parameter to 4 or higher.  If the default dual simplex algorithm seems to be getting stuck, try primal simplex.

It could be a memory issue if there is a lot of memory used to regenerate the problems, so check the memory statistics say with the task manager in Windows or top in Unix, or run a profiler.  Yoiu don't give any details on how you are generating the problems, so it is hard to say.  To see if the generation is the problem, you could write out .sav files and MIP starts at each step and then run the .sav files and MIP starts in the CPLEX Interactive System. 


#CPLEXOptimizers
#DecisionOptimization