Originally posted by: LeonardoJ.

Hi,

i am working on a model in GAMS/CPLEX(11.0), and i have noted after some tests that the root node processing (before b&c) has been very time consuming. Please, see below one of the best cenarios:

*\\*
Total real time on 4 threads = 6.93 sec.
Root relaxation solution time = 6.93 sec.

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

0 0 585.0000 1282 585.0000 15222
0 0 585.0000 87 Cuts: 8 175188
0 0 585.0000 92 Cuts: 97 235853
0 0 585.0000 87 Cuts: 114 291919
0 0 585.0000 97 Cuts: 110 342999
Heuristic still looking.
Heuristic still looking.
Heuristic still looking.
Heuristic still looking.
0 2 585.0000 105 585.0000 397679
Elapsed real time = 1329.61 sec. (tree size = 0.00 MB, solutions = 0)
100 58 611.0000 81 585.0000 2409161
Cuts: 72

• 148 3 integral 0 611.0000 585.0000 2988201 4.26%

Solution limit exceeded.
Starting solution polishing.

152 2 cutoff 611.0000 585.0000 3184026 4.26%
Cuts: 64
153 3 585.0000 264 611.0000 585.0000 3242549 4.26%
Covers: 7
154 3 585.0000 1054 611.0000 585.0000 3308680 4.26%
Cuts: 24

• 155+ 0 585.0000 585.0000 3390049 0.00%

155 0 infeasible 585.0000 585.0000 3399381 0.00%
156 0 589.0000 297 585.0000 585.0000 3436644 0.00%
157 0 cutoff 585.0000 585.0000 3466811 0.00%

GUB cover cuts applied: 28
Cover cuts applied: 162
Zero-half cuts applied: 2

Root node processing (before b&c):
Real time = 1326.84
Parallel b&c, 4 threads:
Real time = 3172.05
Sync time (average) = 355.05
Wait time (average) = 0.00

---

Total (root+branch&cut) = 4498.88 sec.
MIP status(101): integer optimal solution
Fixing integer variables, and solving final LP...
Parallel mode: opportunistic, using up to 4 threads for concurrent optimization.
Tried aggregator 1 time.
LP Presolve eliminated 17985 rows and 10673 columns.
All rows and columns eliminated.

Total real time on 4 threads = 0.09 sec.
Fixed MIP status(1): optimal

Proven optimal solution.

MIP Solution: 585.000000 (3466811 iterations, 158 nodes)
Final Solve: 585.000000 (0 iterations)

Best possible: 585.000000
Absolute gap: 0.000000
Relative gap: 0.000000
*\\*

I have limited the time in 4 hours (14400 seconds) and, for some of the worst instances, GAMS/CPLEX spends the whole time at the root node processing (the branch and bound tree is not started). If i am not wrong, GAMS/CPLEX should only solve the linear relaxation of the problem at the root node (plus adding some cuts) and, although the problem is really difficult, it is not clear for me why it has been so time consuming. Do anyone know some strategy to shorten the time spent at the root node processing??

Thanks a lot and best regards,

Leonardo J.
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

It seems that the resolve of the LP relaxation after each round of cuts is really costing a lot of time. At least the number of simplex iterations spend on each round of cuts is significant.

Additionally, the cuts do not seem to help: the objective value stays constant all the time.

I suggest to either disable cuts completely, or to limit them to a single round (using the CUTPASS parameter).

It may also be the case that primal heuristics consume time. You could try to disable them, but if they do not hurt too much I would rather leave them enabled.
Tobias
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: LeonardoJ.

Hi Tobias,

thank you very much for your tips. I have just performed some tests disabling the cuts, and the time spent at the root node processing decreases well. However, as expected, the following branch and bound has near no effect. I'll still try other strategies...

Best regards,

Leonardo J.
#CPLEXOptimizers
#DecisionOptimization