JuMP and CPLEX - Setting number of threads

vmk97 · July 4, 2022, 3:58pm

I am trying to solve a mixed integer linear program by branch-and-cut using JuMP and CPLEX.

I am having some trouble setting the number of threads for CPLEX.

I am trying to set the parameter CPXPARAM_Threads using set_optimizer_attribute in the following way:

model = Model(CPLEX.Optimizer)
set_optimizer_attribute(model, "CPXPARAM_Threads", 4)

However, in the log I get the following:

Version identifier: 20.1.0.0 | 2020-11-10 | 9bedb6d68
CPXPARAM_Threads                                 1
...
Parallel mode: none, using 1 thread.
...

I don’t understand why the value of 4 that I provided turns into 1.

I would like to run CPLEX in parallel mode with more than one thread.

Thanks in advance.

Obs: I have CPLEX Optimization Studio v20.1 installed which I got through an academic license.
I am running Julia on version 1.7.3, JuMP on version 0.21.10 and CPLEX on version 0.7.8.

gdalle · July 4, 2022, 5:58pm

Not sure what is going on here, but maybe you could try setting this CPLEX parameter too?
Did you launch Julia with a single thread or several?

vmk97 · July 4, 2022, 9:34pm

Thanks, Guillaume.

Yes, I did try to set the CPXPARAM_Parallel parameters as well, but it didn’t work either.

I think the answer is here.

I am passing a custom callback function to CPLEX for handling both lazy constraints and cuts.

According to the link, apparently CPLEX automatically turns off multi-threading when given callbacks.

odow · July 4, 2022, 11:34pm

Did this warning trigger?

github.com

jump-dev/CPLEX.jl/blob/e2f15e06b767b33941dce62873a048d18e6a844f/src/MOI/MOI_callbacks.jl#L101-L108


      
          if MOI.get(model, MOI.NumberOfThreads()) != 1
              @warn(
                  "When using callbacks, make sure to set `NumberOfThreads` to `1` " *
                  "using `MOI.set(model, MOI.NumberOfThreads(), 1)`. Bad things can" *
                  " happen if you don't! Only ignore this message if you know what " *
                  "you are doing."
              )
          end

Perhaps CPLEX started automatically turning off multithreading with callbacks.

But yes, you cannot use threading if you have a callback. The underlying problem is that callbacks from Julia into C are not thread safe, so there is no way to implement this.

vmk97 · July 5, 2022, 12:56pm

First of all, thank you, Guillaume and Oscar.

No, the warning is not triggered. I guess MOI gets the number of threads already forced to one and hence no warning.

Indeed when I run the model with no callback I am able to set the number of threads as I please.

callbacks from Julia into C are not thread safe

Is this specific to Julia? Is it the same for other programming languages?