Pyomo + MPI => 1 environment per process

Answered

Jacob Wren

October 19, 2022 16:09 Edited

• 
• 
• 

Hello,

I am using the third-party modeling framework, Pyomo, as well as MPI (mpi4py) to run several instances of Gurobi in parallel.

How can I have each process create its own environment?

Many thanks,

Jake

0

• 

  Eli Towle

  • Gurobi Staff

  October 21, 2022 18:58 Edited

  Unfortunately, it's not currently possible to explicitly manage Gurobi environments in Pyomo. There is an open issue in the Pyomo GitHub repository about this (Pyomo/pyomo#2408).

  0
• 

  Jacob Wren

  October 21, 2022 19:11

  • 
  • 
  • 

  Hi Eli,

  Is there a workaround to use multiprocessing with Pyomo to ensure that I won't have more than one thread within a single environment?

  Thanks,

  Jake

  0
• 

  Eli Towle

  • Gurobi Staff

  October 21, 2022 21:53

  Do you mean that for each process, you would like to re-use a single environment for all models solved by that process? This would be the best way to limit your token usage. I believe Pyomo creates a new environment for each model. This would explain why you run out of tokens, as you mention in your other community post.

  As far as I know, re-using a single environment like this isn't possible in Pyomo unless you modify Pyomo's source code yourself to support it. The other workaround is to build your model using Gurobi's native Python API.

  0
• 

  Jacob Wren

  October 21, 2022 22:23

  • 
  • 
  • 

  Hi Eli,

  I instead dispose of the environment. This is a fine solution, given the model isn't costly to build.

  Thanks,

  Jake

  0

Please sign in to leave a comment.