Slow MIPS Performance

Ongoing

Animesh Agrawal

November 22, 2020 22:45

• 
• 

Good afternoon, 

My team is experimenting with Gurobi and seeing relatively slow performance on a decently sized model. 

The logs are linked here: https://gist.github.com/AnimeshAgrawal/ad6ef817e1787b09970bc817b8032e72

Our model consists of n binary variables that are multiplied with each other and some constant. The constraints look like (c * x_i * x_j < n) where c and n are constants. As you can see from the logs, the solve performance is quite poor; it seems as though much of the later relaxations run very slowly, but they also appear necessary as some of them do find more optimal solutions (Ex: line 349). 

Does anyone have any suggestions as to how we can improve the model performance? We've experimented with making the following changes: 

m.Params.MIPFocus= 1 
m.Params.Cuts = 3
m.Params.Presolve = 2

But did not see a notable performance improvement. 

0

• Official comment

  

  Simranjit Kaur

  • Gurobi Staff

  December 06, 2024 00:39

  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?.
• 

  Matthias Miltenberger

  • Gurobi Staff

  November 25, 2020 08:57 Edited

  Hi Animesh,

  What makes you think that this is "poor" performance? Just because a model isn't huge doesn't mean that it is easy to solve. From the log, you can see that Gurobi is making steady progress to close the gap after quickly finding a good solution. Often, proving optimality is pretty tough and can result in a large tree even if the optimal solution has been found early.

  For an extreme example, I can recommend the MIPLIB 2017 instance ej. This instance only has 3 variables and one constraint and if you don't solve it in presolving (with Gurobi you need to set Presolving=0), you will build up a very huge tree with very slow progress.

  You might want to look into your model to improve its formulation. Maybe you can find a different way to express the underlying problem and to tighten the LP relaxation.

  Cheers,
  Matthias

  0

Post is closed for comments.