Is there a way to limit the memory used by Gurobi algorithms ?

Answered

Sophie Wuyckens

• June 03, 2022 20:29

Hi,

I am working with big models for which the barrier method seems to be the most approriate solver according to my tuning tests. However, this algorithm uses tremendous memory ressources. Given that I send jobs to a scheduler for my optimizations, I need to specify the memory allocated, number of cpus etc. The problem is that Gurobi can unpredictable peaks in memory and it happens a lot that my jobs are killed because the memory I thought to be sufficient was not. My question therefore is the following: Can we control the maximal memory that guorbi can use to solve the model ? I already know about the parameter NodeFileStart but I can't really use it because it would slower too much the optimization.

 

Thank you in advance.

0

• 

  Jaromił Najman

  Gurobi Staff

  • June 07, 2022 06:30
  • Edited

  Hi Sophie,

  The parameter MemLimit limits the memory used by Gurobi. When the set memory limit is reached, an error is thrown. Thus, this parameter may not be suitable for your application. For very large models it is sometimes just not possible to compress all the model data, i.e., all coefficients, rows, columns, Barrier matrices etc., to a fixed memory amount. I think in your case, you should try to reduce the memory used as discussed in How do I avoid an out-of-memory condition? Usually, the best alternative is to switch to a bigger machine, e.g., a cloud machine, to execute the computations.

  Best regards, 
  Jaromił

  0
• 

  Sophie Wuyckens

  • June 08, 2022 20:15

  

  

  

  Hi Jaromil,

  Thank you very much. Very informative, exactly what I needed to know.

  Regards,

  Sophie

  0

Please sign in to leave a comment.