[SCIP] HELP: Inquiry on Utilizing Multithreading in SCIP via Pyomo

[Turner]

Hi Zhiyu,


I think there has been some misunderstanding on how SCIP and Pyomo interact.

The parameter you mention is related to to the non-standard optimization 
call of SCIP. This can be reached by calling `concurrentopt` instead of 
`optimize` on the SCIP command line interface. What that call does is 
run multiple instances of SCIP in parallel. So it is not a 
parallelisation of a single SCIP call, but rather many different SCIP 
instances (different seeds etc) running in parallel and sharing bound + 
solution information. This feature (maybe someone will correct me) is 
not that well documented or supported. The feature you were probably 
expecting of a parallelised search procedure does not yet exist in SCIP.

The interface we have to Pyomo currently only supports the code behind 
the standard `optimize` call. Therefore, any parameter changes related 
to the `concurrentopt` call do not effect the subsequent optimization 
procedure in your setup. One would need to add the explicit option to 
Pyomo to call SCIPsolveConcurrent for your parameters to make any 
difference. If this feature is truly in demand then I will consider 
adding it to the hopefully soon to exist persistent SCIP interface for 
Pyomo.


Cheers,

Mark


On 22.10.24 13:40, Zhang Zhiyu wrote:
> Dear SCIP,
> I hope this message finds you well. I am currently working on solving 
> a nonlinear programming (NLP) problem using Pyomo and SCIP, and I am 
> seeking guidance on effectively utilizing multithreading to improve 
> solve times.
> Here is the setup I am using:
>
>  *
>     Installed Pyomo version 6.8.0 and SCIP version 9.1.0 via conda.
>  *
>     Created the solver in Pyomo using:
>      o
>         solver = pyo.SolverFactory('scip')
>  *
>     Set the multithreading option:
>      o
>         solver.options['parallel/maxnthreads'] = 16
>
> Despite setting the number of threads to 16, I have observed that the 
> solve time does not seem to improve. I would greatly appreciate any 
> insights or recommendations you might have for enabling efficient 
> multithreading with SCIP in this context.
>
> Thank you for your time and assistance.
> Best regards,
> Zhiyu Zhang
> zhidaoxiayu_Biu at outlook.com
>
>
> _______________________________________________
> Scip mailing list
> Scip at zib.de
> https://listserv.zib.de/mailman/listinfo/scip
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.zib.de/pipermail/scip/attachments/20241022/72b53a9b/attachment.html>