[SCIP] FiberSCIP is very slow in deterministic mode

[Yuji Shinano]

Dear Vladimir V. Voloshinov,

On 14.03.25 16:36, Vladimir V. Voloshinov wrote:
> Dear SCIP team,
> wehave been successfullyusingSCIP & FiberSCIP & ParaSCIPfor a longtime.
> But sometimes some problems arise...
> Recently we found that FiberSCIP works very very slowly in deterministic mode.
> Moreover, it looks like */Deterministic=TRUE/* (in ug.set file) switches off multithreading and regardless of */-sth/* setting total CPU load becomes the same as for the case */-sth=1/*.
> But performance becomes even worse than for one threaded SCIP... See example below.
>
> The question is: maybe there are some other ways to get a deterministic behaviour of FiberSCIP keeping performance? The question arose when we tried to compare FiberSCIP performance and results on different computing environments...

No, unfortunately. The deterministic mode was developed only to debug FiberSCIP more efficiently.
Please read the FiberSCIP paper:
https://pubsonline.informs.org/doi/10.1287/ijoc.2017.0762
And, it would not always be deterministic with the current version of SCIP.

Best regards,
Yuji

>
> Below is an example of results for some NLP global optimization problem without discrete variables but with SOS2 constraints:
> ___________________________________________________________
> Presolved Problem  :
>   Variables       : 145 (0 binary, 0 integer, 0 implicit integer, 145 continuous)
>   Constraints    : 61
> Constraints      : Number
>   SOS2             : 15
>   linear              : 30
>   nonlinear        : 16
> --------------------------------------------------------------------------------------------------
> It was solved with total 10% gap, */-sth=16/* and */Deterministic=FALSE/* in 87 sec:
> _____________________________________________________
> SCIP Status        : solving was interrupted [given gap reached]
> Total Time         : 86.80
> B&B Tree           :   nodes (total)    : 202112
> Solution           :
>   Primal Bound     : +3.10185241045302e-03
>   Dual Bound       : +2.98514588645749e-03
> Gap                : 3.90958 %
> ----------------------------------------------------------------------------------------
>
> But solving time with the same settings and /*Deterministic=TRUE*/ became almost 50 times bigger:
> _____________________________________________________
> SCIP Status        : solving was interrupted [given gap reached]
> Total Time         : 4216.07
> B&B Tree           :   nodes (total)    : 141474
> Solution           :
>   Primal Bound     : +3.09812857597093e-03
>   Dual Bound       : +2.98180690970866e-03
> Gap                : 3.90105 %
> ----------------------------------------------------------------------------------------
>
> Sincerely yours,
> Vladimir V. Voloshinov,
> web: GoogleScholar profile <https://scholar.google.ru/citations?hl=en&user=-m4QhNEAAAAJ&view_op=list_works&sortby=pubdate>
>
> _______________________________________________
> Scip mailing list
> Scip at zib.de
> https://listserv.zib.de/mailman/listinfo/scip


-- 
--------------------------------------------------
Yuji Shinano
Zuse Institute Berlin
Takustrasse 7, 14195 Berlin-Dahlem, Germany
Phone:+49 30 84185-477, Fax:+49 30 84185-269
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.zib.de/pipermail/scip/attachments/20250314/adfd2ba4/attachment.html>