The time to explore nodes fluctuates using different number of threads
AnsweredHi,
I have been solving the same problem by setting different number of threads. The time fluctuates, for example, the minimum time is 228s by using 6 threads and the maximum time is 563 seconds by using 10 threads. Is it because "For MIPs that are solved at or near the root node, more threads generally will not help much."? Thanks for your time.
Explored 1 nodes (134170 simplex iterations) in 296.53 seconds
Thread count was 1 (of 12 available processors)
Solution count 9: 338.5 338.5 338.5 ... 810.64
Explored 1 nodes (104204 simplex iterations) in 298.86 seconds
Thread count was 2 (of 12 available processors)
Solution count 9: 338.5 338.5 411.36 ... 810.64
Explored 1 nodes (99781 simplex iterations) in 270.34 seconds
Thread count was 3 (of 12 available processors)
Solution count 6: 338.5 338.5 411.36 ... 573.04
Explored 1 nodes (214606 simplex iterations) in 319.40 seconds
Thread count was 4 (of 12 available processors)
Solution count 8: 338.5 338.5 351.4 ... 614.8
Explored 1 nodes (212865 simplex iterations) in 322.99 seconds
Thread count was 5 (of 12 available processors)
Solution count 10: 338.5 338.5 338.5 ... 614.8
Explored 1 nodes (97701 simplex iterations) in 228.50 seconds
Thread count was 6 (of 12 available processors)
Solution count 8: 338.5 338.5 424.26 ... 661.86
Explored 1 nodes (112467 simplex iterations) in 276.43 seconds
Thread count was 7 (of 12 available processors)
Solution count 7: 338.5 338.5 471.32 ... 661.86
Explored 1 nodes (223162 simplex iterations) in 319.99 seconds
Thread count was 8 (of 12 available processors)
Solution count 10: 338.5 338.5 351.4 ... 661.86
Explored 1 nodes (115531 simplex iterations) in 268.50 seconds
Thread count was 9 (of 12 available processors)
Solution count 6: 338.5 338.5 424.26 ... 661.86
Explored 1 nodes (254407 simplex iterations) in 563.10 seconds
Thread count was 10 (of 12 available processors)
Solution count 10: 338.5 338.5 364.3 ... 661.86
Explored 1 nodes (118078 simplex iterations) in 381.13 seconds
Thread count was 11 (of 12 available processors)
Solution count 8: 338.5 338.5 437.16 ... 661.86
Explored 1 nodes (223162 simplex iterations) in 467.41 seconds
Thread count was 12 (of 12 available processors)
Solution count 10: 338.5 338.5 351.4 ... 661.86
-
Is it because "For MIPs that are solved at or near the root node, more threads generally will not help much."?
Yes, this can be one reason. Another reason for the time fluctuation is that changing the number of Threads used to solve a problem is not deterministic, cf. Is Gurobi Optimizer deterministic? Any change to a parameter (including the number of threads) can change the solution path significantly and result in different performance. This is known as Performance Variability.
Best regards,
Jaromił0
Please sign in to leave a comment.
Comments
1 comment