When I run .NET 4 application which uses CPLEX, I get different outputs on different machines. On my development machine CPLEX outputs one result (which is abnormal and stuck at some large value), and on all other machines, result is OK.
First, I thought that it has something to do with OS, and since I have both Windows 7 x64 and Windows 8 x64 on my development machine, I tried running application on both systems. Result is the same - faulty.
Then I tried running on two different desktop machines, and it works OK. I even tried inside virtual machine, with Windows 7 guest on my development machine, and in all those cases, result is OK.
How to fix this issue? How can I debug what is going on? Any suggestion would be helpful.
Additional info:
Here is the faulty output from my development machine:
IBM ILOG CPLEX Optimization Studio Commercial Edition
Tried aggregator 1 time.
MIP Presolve eliminated 125 rows and 0 columns.
MIP Presolve modified 15000 coefficients.
Reduced MIP has 28795 rows, 14520 columns, and 965648 nonzeros.
Reduced MIP has 14520 binaries, 0 generals, 0 SOSs, and 0 indicators.
Probing time = 0.05 sec.
Tried aggregator 1 time.
Presolve time = 1.39 sec.
Found feasible solution after 1.48 sec. Objective = 4.5736e+016
Probing time = 0.03 sec.
Clique table members: 14520.
MIP emphasis: balance optimality and feasibility.
MIP search method: dynamic search.
Parallel mode: deterministic, using up to 8 threads.
Root relaxation solution time = 2.70 sec.
Nodes Cuts/
Node Left Objective IInf Best Integer Best Bound ItCnt Gap
* 0+ 0 4.57357e+016 1870 ---
* 0+ 0 2.93779e+016 1870 ---
0 0 1.54596e+015 811 2.93779e+016 1.54596e+015 1870 94.74%
* 0+ 0 5.78471e+015 1.54596e+015 2258 73.28%
0 0 1.59388e+015 942 5.78471e+015 Cuts: 377 2258 72.45%
0 0 1.65120e+015 813 5.78471e+015 ZeroHalf: 216 2590 71.46%
0 0 1.67178e+015 968 5.78471e+015 ZeroHalf: 285 2857 71.10%
0 0 1.71285e+015 957 5.78471e+015 ZeroHalf: 123 3241 70.39%
0 0 1.73687e+015 991 5.78471e+015 ZeroHalf: 148 3642 69.97%
0 0 1.76181e+015 997 5.78471e+015 ZeroHalf: 50 4048 69.54%
0 0 1.77762e+015 976 5.78471e+015 ZeroHalf: 23 4344 69.27%
0 0 1.78664e+015 977 5.78471e+015 ZeroHalf: 22 4601 69.11%
0 0 1.80055e+015 994 5.78471e+015 ZeroHalf: 75 4943 68.87%
0 0 1.80615e+015 993 5.78471e+015 ZeroHalf: 9 5143 68.78%
0 0 1.81464e+015 996 5.78471e+015 ZeroHalf: 33 5440 68.63%
0 0 1.82294e+015 991 5.78471e+015 Cuts: 27 5765 68.49%
0 0 1.83208e+015 989 5.78471e+015 Cuts: 18 6104 68.33%
0 0 1.84944e+015 946 5.78471e+015 Cuts: 182 6698 68.03%
0 0 1.88828e+015 956 5.78471e+015 Cuts: 269 7432 67.36%
0 0 1.89984e+015 971 5.78471e+015 Cuts: 118 7822 67.16%
0 0 1.91068e+015 985 5.78471e+015 Cuts: 45 8155 66.97%
0 0 1.92274e+015 980 5.78471e+015 Cuts: 40 8597 66.76%
0 0 1.94969e+015 977 5.78471e+015 Cuts: 121 9295 66.30%
0 0 1.95749e+015 970 5.78471e+015 Cuts: 69 9682 66.16%
0 0 1.96689e+015 978 5.78471e+015 Cuts: 42 10093 66.00%
0 0 1.97224e+015 980 5.78471e+015 Cuts: 30 10392 65.91%
0 0 1.97649e+015 978 5.78471e+015 Cuts: 13 10622 65.83%
0 0 1.98557e+015 976 5.78471e+015 Cuts: 50 10876 65.68%
0 0 1.99259e+015 987 5.78471e+015 Cuts: 46 11183 65.55%
0 0 2.00116e+015 987 5.78471e+015 Cuts: 77 11494 65.41%
0 0 2.01149e+015 990 5.78471e+015 Cuts: 37 11921 65.23%
0 0 2.01969e+015 976 5.78471e+015 Cuts: 60 12286 65.09%
0 0 2.02492e+015 966 5.78471e+015 Cuts: 43 12578 65.00%
0 0 2.03316e+015 969 5.78471e+015 Cuts: 84 12881 64.85%
0 0 2.04175e+015 971 5.78471e+015 Cuts: 52 13174 64.70%
0 0 2.05273e+015 964 5.78471e+015 Cuts: 68 13649 64.51%
0 0 2.05591e+015 966 5.78471e+015 Cuts: 23 13902 64.46%
0 0 2.06170e+015 964 5.78471e+015 Cuts: 43 14235 64.36%
0 0 2.06574e+015 971 5.78471e+015 Cuts: 46 14481 64.29%
0 0 2.07500e+015 952 5.78471e+015 Cuts: 47 14890 64.13%
0 0 2.08393e+015 941 5.78471e+015 Cuts: 53 15280 63.98%
0 0 2.09065e+015 948 5.78471e+015 Cuts: 36 15601 63.86%
0 0 2.09803e+015 955 5.78471e+015 Cuts: 104 16015 63.73%
0 0 2.10316e+015 944 5.78471e+015 Cuts: 63 16384 63.64%
0 0 2.11460e+015 948 5.78471e+015 Cuts: 76 16846 63.45%
0 0 2.12942e+015 956 5.78471e+015 Cuts: 97 17366 63.19%
0 0 2.16151e+015 931 5.78471e+015 Cuts: 327 18262 62.63%
0 0 2.17502e+015 930 5.78471e+015 Cuts: 120 18714 62.40%
0 0 2.27173e+015 884 5.78471e+015 Cuts: 385 20411 60.73%
0 0 2.29553e+015 899 5.78471e+015 Cuts: 134 21117 60.32%
0 0 2.30977e+015 920 5.78471e+015 Cuts: 71 21621 60.07%
0 0 2.31979e+015 909 5.78471e+015 Cuts: 38 22064 59.90%
0 0 2.32293e+015 911 5.78471e+015 Cuts: 33 22333 59.84%
0 0 2.32804e+015 908 5.78471e+015 Cuts: 60 22682 59.76%
0 0 2.33324e+015 906 5.78471e+015 Cuts: 43 22991 59.67%
0 0 2.34725e+015 909 5.78471e+015 Cuts: 22 23332 59.42%
0 0 2.35327e+015 903 5.78471e+015 Cuts: 72 23571 59.32%
0 0 2.36485e+015 889 5.78471e+015 Cuts: 61 23938 59.12%
0 0 2.37310e+015 884 5.78471e+015 Cuts: 45 24319 58.98%
0 0 2.39039e+015 885 5.78471e+015 Cuts: 79 24958 58.68%
0 0 2.39677e+015 901 5.78471e+015 Cuts: 56 25186 58.57%
0 0 2.40149e+015 885 5.78471e+015 Cuts: 41 25442 58.49%
0 0 2.40325e+015 883 5.78471e+015 Cuts: 60 25617 58.46%
0 0 2.41812e+015 894 5.78471e+015 Cuts: 84 26070 58.20%
0 0 2.42607e+015 903 5.78471e+015 Cuts: 68 26474 58.06%
0 0 2.52725e+015 844 5.78471e+015 Cuts: 524 28147 56.31%
0 0 2.55516e+015 835 5.78471e+015 Cuts: 196 28727 55.83%
0 0 3.23302e+015 654 5.78471e+015 Cuts: 889 30777 44.11%
0 0 3.32795e+015 604 5.78471e+015 Cuts: 337 31390 42.47%
0 0 3.37780e+015 686 5.78471e+015 Cuts: 212 31791 41.61%
0 0 3.41202e+015 701 5.78471e+015 Cuts: 121 32148 41.02%
0 0 3.52260e+015 654 5.78471e+015 Cuts: 398 33114 39.10%
0 0 3.57604e+015 639 5.78471e+015 Cuts: 244 33613 38.18%
0 0 3.61288e+015 625 5.78471e+015 Cuts: 235 34098 37.54%
0 0 3.67035e+015 615 5.78471e+015 Cuts: 163 34491 36.55%
0 0 3.73087e+015 663 5.78471e+015 Cuts: 157 35105 35.50%
0 0 3.75834e+015 668 5.78471e+015 Cuts: 144 35530 35.03%
0 0 3.77753e+015 662 5.78471e+015 Cuts: 93 35899 34.70%
0 0 3.81446e+015 660 5.78471e+015 Cuts: 82 36310 34.06%
0 0 3.83284e+015 648 5.78471e+015 Cuts: 76 36608 33.74%
0 0 3.84779e+015 656 5.78471e+015 Cuts: 58 36845 33.48%
0 0 3.85941e+015 639 5.78471e+015 Cuts: 67 37084 33.28%
0 0 3.87033e+015 649 5.78471e+015 Cuts: 40 37284 33.09%
0 0 3.87967e+015 662 5.78471e+015 Cuts: 32 37422 32.93%
0 0 3.88305e+015 666 5.78471e+015 Cuts: 24 37577 32.87%
0 0 3.89084e+015 654 5.78471e+015 Cuts: 59 37771 32.74%
0 0 3.89489e+015 654 5.78471e+015 Cuts: 38 37968 32.67%
0 0 3.90160e+015 669 5.78471e+015 Cuts: 55 38172 32.55%
0 0 3.90440e+015 663 5.78471e+015 Cuts: 40 38329 32.50%
0 0 3.90794e+015 660 5.78471e+015 Cuts: 27 38492 32.44%
0 0 3.91249e+015 661 5.78471e+015 Cuts: 31 38662 32.36%
0 0 3.91569e+015 673 5.78471e+015 Cuts: 32 38814 32.31%
0 2 3.91569e+015 673 5.78471e+015 3.91569e+015 38814 32.31%
Elapsed real time = 312.08 sec. (tree size = 0.01 MB, solutions = 3)
1 3 4.19433e+015 728 5.78471e+015 3.91569e+015 41427 32.31%
2 2 cutoff 5.78471e+015 3.91569e+015 43784 32.31%
3 3 4.35193e+015 730 5.78471e+015 3.91569e+015 45337 32.31%
4 2 cutoff 5.78471e+015 3.91569e+015 46000 32.31%
5 3 4.52240e+015 741 5.78471e+015 3.91569e+015 47554 32.31%
6 2 cutoff 5.78471e+015 3.91569e+015 49263 32.31%
7 3 4.66098e+015 799 5.78471e+015 3.91569e+015 51056 32.31%
8 2 cutoff 5.78471e+015 3.91569e+015 51943 32.31%
9 3 4.85025e+015 807 5.78471e+015 3.91569e+015 52422 32.31%
12 2 cutoff 5.78471e+015 3.91569e+015 53363 32.31%
Elapsed real time = 350.88 sec. (tree size = 0.01 MB, solutions = 3)
15 3 5.59278e+015 896 5.78471e+015 3.91569e+015 54625 32.31%
Clique cuts applied: 1901
Implied bound cuts applied: 160
Zero-half cuts applied: 150
Root node processing (before b&c):
Real time = 310.11
Parallel b&c, 8 threads:
Real time = 39.39
Sync time (average) = 0.00
Wait time (average) = 0.00
-------
Total (root+branch&cut) = 349.50 sec.
inst\instance.txt 00:05:50.9629598 5,78471010310452E+15
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1 0 0 0 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 0 1 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 1 1 1 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 0 0 0 0
And here is the normal output from all other machines:
IBM ILOG CPLEX Optimization Studio Commercial Edition
Tried aggregator 1 time.
MIP Presolve eliminated 129 rows and 0 columns.
MIP Presolve modified 15480 coefficients.
Reduced MIP has 28791 rows, 14520 columns, and 965163 nonzeros.
Reduced MIP has 14520 binaries, 0 generals, 0 SOSs, and 0 indicators.
Probing time = 0.03 sec.
Tried aggregator 1 time.
Presolve time = 1.51 sec.
Found feasible solution after 1.62 sec. Objective = 5170425.9868
Probing time = 0.05 sec.
Clique table members: 14520.
MIP emphasis: balance optimality and feasibility.
MIP search method: dynamic search.
Parallel mode: deterministic, using up to 8 threads.
Root relaxation solution time = 4.82 sec.
Nodes Cuts/
Node Left Objective IInf Best Integer Best Bound ItCnt Gap
* 0+ 0 5170425.9868 2742 ---
* 0+ 0 2176197.5073 2742 ---
0 0 963286.8006 1320 2176197.5073 963286.8006 2742 55.74%
* 0+ 0 1742629.0387 963286.8006 3146 44.72%
0 0 968918.1899 1292 1742629.0387 Cuts: 180 3146 44.40%
0 0 972621.9253 1285 1742629.0387 ZeroHalf: 36 3543 44.19%
0 0 974930.7260 1336 1742629.0387 ZeroHalf: 157 3934 44.05%
0 0 975330.5044 1344 1742629.0387 ZeroHalf: 9 4042 44.03%
0 0 976752.3324 1344 1742629.0387 ZeroHalf: 23 4298 43.95%
0 0 977523.1431 1375 1742629.0387 ZeroHalf: 13 4523 43.91%
0 0 978030.2843 1358 1742629.0387 ZeroHalf: 7 4732 43.88%
0 0 978582.8390 1290 1742629.0387 Cuts: 13 4897 43.84%
0 0 979462.4757 1394 1742629.0387 Cuts: 22 5107 43.79%
0 0 980081.0140 1361 1742629.0387 Cuts: 16 5256 43.76%
0 0 986296.0889 1453 1742629.0387 Cuts: 135 6066 43.40%
0 0 987753.3557 1412 1742629.0387 Cuts: 105 6386 43.32%
0 0 988885.7034 1442 1742629.0387 Cuts: 57 6694 43.25%
0 0 989742.4516 1434 1742629.0387 Cuts: 27 6924 43.20%
0 0 990497.2364 1434 1742629.0387 Cuts: 45 7178 43.16%
* 0+ 0 1402648.1251 990497.2364 7489 29.38%
0 0 991849.2315 1437 1402648.1251 Cuts: 288 7489 29.29%
0 0 1243733.1371 593 1402648.1251 Cuts: 2446 12636 11.33%
0 0 1276887.2546 583 1402648.1251 Cuts: 556 14078 8.97%
0 0 1296511.1088 701 1402648.1251 Cuts: 600 14813 7.57%
0 0 1305221.3720 679 1402648.1251 Cuts: 193 15488 6.95%
0 0 1309747.0795 660 1402648.1251 Cuts: 139 15848 6.62%
0 0 1318085.1042 725 1402648.1251 Cuts: 43 16281 6.03%
0 0 1323312.2600 747 1402648.1251 Cuts: 55 16728 5.66%
0 0 1325171.8791 677 1402648.1251 Cuts: 23 17033 5.52%
0 0 1326274.8424 704 1402648.1251 Cuts: 19 17265 5.44%
0 0 1327853.0467 714 1402648.1251 Cuts: 20 17519 5.33%
0 0 1328078.7180 687 1402648.1251 Cuts: 6 17621 5.32%
0 0 1328705.5414 702 1402648.1251 Cuts: 18 17768 5.27%
0 0 1329753.6276 716 1402648.1251 Cuts: 15 17988 5.20%
0 0 1330728.0832 705 1402648.1251 ZeroHalf: 5 18165 5.13%
0 0 1331208.9314 707 1402648.1251 Cuts: 15 18322 5.09%
0 0 1331935.6017 709 1402648.1251 ZeroHalf: 5 18432 5.04%
0 0 1332438.9391 738 1402648.1251 ZeroHalf: 4 18536 5.01%
0 0 1332565.6221 712 1402648.1251 Cuts: 3 18578 5.00%
0 0 1332961.2175 712 1402648.1251 ZeroHalf: 7 18653 4.97%
0 2 1332962.3523 711 1402648.1251 1332962.3523 18656 4.97%
Elapsed real time = 136.63 sec. (tree size = 0.01 MB, solutions = 4)
1 3 1355937.1387 819 1402648.1251 1332962.3523 20475 4.97%
2 2 cutoff 1402648.1251 1332962.3523 20810 4.97%
3 3 1373864.6210 889 1402648.1251 1332962.3523 21989 4.97%
5 3 1391805.4913 902 1402648.1251 1332962.3523 22812 4.97%
6 2 cutoff 1402648.1251 1332962.3523 24410 4.97%
7 1 cutoff 1402648.1251 1332962.3523 25377 4.97%
8 0 cutoff 1402648.1251 1332962.3523 26154 4.97%
Clique cuts applied: 2880
Implied bound cuts applied: 73
Zero-half cuts applied: 97
Root node processing (before b&c):
Real time = 134.80
Parallel b&c, 8 threads:
Real time = 16.44
Sync time (average) = 0.00
Wait time (average) = 0.00
-------
Total (root+branch&cut) = 151.24 sec.
inst\instance.txt 00:02:32.8137404 1402648.12514155
0 0 0 0 1 0 0 0 0 0 0 1 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
First of all, you should be absolutely sure that you're solving exactly the same problem.
Looking at the logs, it seems that the problem CPLEX is solving is different.
For example presolve eliminates 125 rows in the first case and 129 in the second, and modifies 15000 coefficients in the first case and 15480 in the second; also the reduced MIP seems different:
28795 rows, 14520 columns, and 965648 nonzeros in the first case and 28791 rows, 14520 columns, and 965163 nonzeros in the second case.
I'm pretty sure CPLEX presolve is totally deterministic and definitely platform independent. So in my opinion, for some reason you're generating two different problems.
You could try to export the problem to LP file on dev and test machines, then check if the two files are equal (or even try to solve the .LP file generated on test machine to the development one and check the log)
Anyway, on IBM site there's a section about "runs reproducibility".
They state that on different HW architectures results can actually change, but optimal solution objective value (of course) remains the same.
However, I really doubt this is your case (values are too different), so I'd compare the exported LP problems to see if there are any difference between development and test machines.