Originally posted by: MichaelD.Moffitt

Consider the following model:

http://www.eecs.umich.edu/~mmoffitt/model_56.lp

and program execution (using the x86-64_debian4.0_4.1 binary):

CPLEX> read model_56.lp
Problem 'model_56.lp' read.
Read time = 0.01 sec.
CPLEX> mipopt

From all appearances, CPLEX enters into an infinite loop around node 21500:

Elapsed real time = 8.83 sec. (tree size = 1.02 MB, solutions = 12)
21100 2707 cutoff 316.0000 312.0000 242735 1.27%
21200 2726 312.0000 14 316.0000 312.0000 244236 1.27%
21300 2743 312.9582 31 316.0000 312.0000 246156 1.27%
21400 2750 infeasible 316.0000 312.0000 247872 1.27%
21500 2793 cutoff 316.0000 312.0000 248989 1.27%
hangs for hours

Any thoughts? Can others reproduce this?
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

I was able to reproduce your issue, thanks for the bug report.

On my machine (2 core x86-64 linux) the solve process hangs after about 41000 nodes. Using a debug build of CPLEX, I get a floating exception after 28600 nodes.

Your model does not look very ugly in terms of numerics, with the exception of pretty large upper bounds for the range variables. Maybe, this is causing the trouble.
Tobias
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

The issue is inside the simplex algorithm. I was able to work around the problem by setting the pricing strategy to steepest edge:

set simplex dgradient 4

This may fix the problem also on your machine, but it could be just luck.
We will investigate the issue and hopefully have it fixed in the next release.
Tobias
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

Here are the concluding remarks for this bug:

1. It's known and fixed in the next release.
2. This is extremely rare.
3. Because it is so rare, a change like switching to a different pricing strategy
is likely to provide a robust work around until the next release becomes available.
Cheers,

Tobias
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: MichaelD.Moffitt

Great, thanks for looking into this so quickly. I'll post a follow-up if I run into any problems with your proposed change.
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

You can't always draw conclusions from one sample model, but I'll mention that I noticed that the RINS heuristic does a terrific job on this model if you call it frequently. "Set mip str rins 100" or even smaller worked quite well for me. You might try it and see whether it helps on other similar models you run.

By the way, do you mind if we keep the model you posted in our model library for future testing and validation purposes?
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: MichaelD.Moffitt

Thanks for the tip, I'll try RINS in the future.

Sure, feel free to keep this model. I actually hit this bug one other time (out of several thousand runs) ... if I can regenerate that instance I'll send it along as well.
#CPLEXOptimizers
#DecisionOptimization