Very slow MIP solving for small problem, how to speed up?
Hello,
my model is being solved extremely slowly, and I believe it's because of the Markowitz tolerance (though I am by no means an expert). Here's the relevant logs:
2020-02-06 15:39:13,084 DEBUG: Model has 1 quadratic constraint
2020-02-06 15:39:13,084 DEBUG: Variable types: 134 continuous, 80685 integer (80685 binary)
2020-02-06 15:39:13,086 DEBUG: Coefficient statistics:
2020-02-06 15:39:13,086 DEBUG: Matrix range [1e-03, 2e+01]
2020-02-06 15:39:13,086 DEBUG: QMatrix range [1e+00, 1e+00]
2020-02-06 15:39:13,086 DEBUG: QLMatrix range [1e+00, 1e+00]
2020-02-06 15:39:13,086 DEBUG: Objective range [1e+00, 1e+00]
2020-02-06 15:39:13,086 DEBUG: Bounds range [1e+00, 1e+00]
2020-02-06 15:39:13,087 DEBUG: RHS range [1e-03, 3e+01]
2020-02-06 15:39:16,969 DEBUG: Presolve removed 1412 rows and 741 columns
2020-02-06 15:39:16,970 DEBUG: Presolve time: 3.87s
2020-02-06 15:39:16,971 DEBUG: Presolved: 38737 rows, 80083 columns, 891597 nonzeros
2020-02-06 15:39:16,971 DEBUG: Presolved model has 5 bilinear constraint(s)
2020-02-06 15:39:17,177 DEBUG: Variable types: 16 continuous, 80067 integer (80006 binary)
2020-02-06 15:39:17,207 DEBUG:
2020-02-06 15:39:17,207 DEBUG: Deterministic concurrent LP optimizer: primal and dual simplex
2020-02-06 15:39:17,207 DEBUG: Showing first log only...
2020-02-06 15:39:17,207 DEBUG:
2020-02-06 15:39:18,071 DEBUG: Warning: Markowitz tolerance tightened to 0.5
The root relaxation was solved reasonably quickly (110 seconds and 62558 iterations), but after the weekend only 15945 nodes were explored in 320524 seconds, each node taking a few thousands iterations. This on a 20-cores machine. I feel like this is very slow considering the size of the instance.
What could be the reason for this? Which parameters should I try to tune?
-
Official comment
This post is more than three years old. Some information may not be up to date. For current information, please check the Gurobi Documentation or Knowledge Base. If you need more help, please create a new post in the community forum. Or why not try our AI Gurobot?. -
You can try experimenting with NodeMethod parameter,
https://www.gurobi.com/documentation/9.0/refman/nodemethod.html
as sometimes non-default setting (e.g., barrier) does better for certain models.
Also, seeing a more complete log from your run in the post could be helpful; without much information it is difficult to make any recommendations.
Hope this helps.
0
Post is closed for comments.
Comments
2 comments