Originally posted by: bockca78

Dear CPLEX community,

I am tryping to solve a MIP problem with the following characteristics with CPLEX 12.2.0.0/GAMS on Windows Server 2003, Enterprise x64, Service Pack 2 machine with 65 GB ram.

• Beginning of log file:

  ---

• E2E-SCOpt_real_V5_no_heuristic_NB_LF.gms(9835215) 31094 Mb
• 10,232,736 rows 9,885,259 columns 52,715,948 non-zeroes
• 55,758 discrete-columns

IBM ILOG CPLEX Aug 18, 2010 23.5.2 WEX 19143.19383 WEI x86_64/MS Windows
Cplex 12.2.0.0, GAMS Link 34
GAMS/Cplex licensed for continuous and discrete problems.

Reading parameter(s) from "cplex.opt"
>> workdir=e:\tmp_CPLEX
>> threads=1
>> *clonelogs=1
>> epgap=0.1
>> *epint=0
>> solvefinal=0
>> memoryemphasis=1
>> mipemphasis=1
>> mipinterval 1
>> varsel 4
>> cuts no
>> *heurfreq -1
>> heurfreq 5
>> rinsheur 5

Finished reading from "cplex.opt"
Reading data...
Starting Cplex...
Presolve has eliminated 130458 rows and 1989148 columns...
Presolve has eliminated 2110671 rows and 1994475 columns...
Presolve has eliminated 2110671 rows and 1994475 columns...
Presolve has eliminated 2116304 rows and 1994483 columns...
Aggregator has done 219 substitutions...
Tried aggregator 2 times.
MIP Presolve eliminated 2116304 rows and 1995920 columns.
MIP Presolve modified 2 coefficients.
Aggregator did 219 substitutions.
Reduced MIP has 8116213 rows, 7889120 columns, and 31346541 nonzeros.
Reduced MIP has 35082 binaries, 0 generals, 0 SOSs, and 0 indicators.
Elapsed time 10.02 sec. for 49% of probing (26 vars fixed).
Probing fixed 38 vars, tightened 7866 bounds.
Probing time = 13.45 sec.
Presolve has eliminated 58292 rows and 187779 columns...
Presolve has eliminated 88641 rows and 188164 columns...
Presolve has eliminated 88738 rows and 188261 columns...
Tried aggregator 1 time.
Presolve has eliminated 88804 rows and 188327 columns...
MIP Presolve eliminated 88804 rows and 188327 columns.
MIP Presolve modified 22356 coefficients.
Reduced MIP has 8027409 rows, 7700793 columns, and 30860697 nonzeros.
Reduced MIP has 30524 binaries, 0 generals, 0 SOSs, and 0 indicators.
Presolve time = 146.20 sec.
Found feasible solution after 154.91 sec. Objective = 4.3740e+008
Probing time = 3.95 sec.
Clique table members: 5635.
MIP emphasis: integer feasibility.
MIP search method: dynamic search.
Parallel mode: none, using 1 thread.
Initializing dual steep norms . . .

Iteration log . . .
... snipp
Elapsed time = 1589.69 sec. (551000 iterations).
Iteration: 551404 Dual objective = 231142475.642953
Removing perturbation.
Iteration: 551482 Scaled dual infeas = 0.000000
Iteration: 551485 Dual objective = 231142475.176589
Root relaxation solution time = 1609.30 sec.

Nodes Cuts/
Node Left Objective IInf Best Integer Best Node ItCnt Gap

• 0+ 0 4.37404e+008 551487 ---

0 0 2.31142e+008 12965 4.37404e+008 2.31142e+008 551487 47.16%
Heuristic still looking.
Heuristic still looking.
Heuristic still looking.
Heuristic still looking.

• Beginning of log file:

  ---

CPLEX never comes back from the heuristics on this problem. Will I have to state the MIP as infeasible to our client or is there any chance to come up with an integer feasible solution with <47% gap?

Thanks so much for you help,
Carsten
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

Actually, we worked pretty hard for CPLEX 12.2 to prevent cases like this where the heuristics just run into an almost infinite loop. So, you seem to have found a case that we have missed.

It would be great if you could send me your model such that I can see what is going on. I guess this is way too large for an email attachment. If you are allowed to send us your model, I can give you instructions on how to upload it. Don't upload files of this size to the forum.

But maybe this is just a bug in CPLEX 12.2. Did you already try to use CPLEX 12.2.0.2 on your model?

In any case, you can of course disable the heuristics (in the interactive with "set mip str heur -1"). But this also means that you would not get your initial 47% gap solution and you have to hope that the "integer" part of the model is not too complicated such that CPLEX will find a solution in the tree in a few number of nodes.
Tobias
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: bockca78

Hello Tobias,
thanks for your quick response and your offer on further examining my problem. I have to check on providing the model, but what would be the best way to provide you with the model. I have GAMS code and a lot of *.inc files to feed data to the model. Will you be able to get this running in your environment?

Best regards,
Carsten
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

I doubt if Tobias has access to GAMS, so analysis of the model outside the GAMS environment will be necessary. GAMS can export a model into the industry-standard MPS format(human readable ASCII format). I don't have my GAMS manual handy, but you can have a look at the following URL for information:
http://interfaces.gams-software.com/doku.php?id=mps-files

The MPS file will probably be huge, so if you can compress it with gzip that will be helpful before you upload it (and Tobias downloads it). CPLEX can read gzipped MPS files directly, another inducement to investing the time to compress first.
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

Hi Carsten,

please follow John's suggestion and convert your GAMS model into an .mps (or even better a .sav) file, and then compress it (preferably with gzip or zip).

Then upload it via FTP as follows:

ftp testcase.boulder.ibm.com
# username = anonymous, password = <your email address>
cd /software/toibm/ilog
bin
put <filename>
quit

Then post the name of your file here.
Tobias
#CPLEXOptimizers
#DecisionOptimization