[Scip] How to activate/deactivate parallel mode in CPLEX
Ambros Gleixner
gleixner at zib.de
Wed Jul 16 17:14:30 CEST 2014
Hi Luciano,
sorry, my first answer was incomplete. The default value of the
lp/threads parameter is already 0, which means "automatic", i.e., the LP
solver can choose how many threads it wants to use. I expect that Cplex
will try to exploit all cores in this setting, so there is no need to
change this.
The issue why you are not seeing them used is because by default we
explictly call the (primal or dual) simplex method of Cplex, which is
not parallel. The choice of LP algorithm in SCIP is governed by the two
parameters
lp/initalgorithm and lp/resolvealgorithm
If you set these to "b" or "c", SCIP will call the barrier method of
Cplex, which exploits parallelism. ("c" is recommended because it will
provide a basic LP solution, which is needed for the generation of
tableau cutting planes.)
Cplex also offers the possibility to call simplex and barrier
concurrently. However, the current version of SCIP does not interface
to this. We plan to provide this functionality soon. But for now you
should experimentally figure out whether for your instances the barrier
is better than the simplex and set the above parameters accordingly.
Hope this works for you,
Ambros
Am 15.07.2014 18:48, schrieb Luciano Porretta:
> What if does not work? ;)
>
> --
> -----------------------------------------------------------------------------------------
> PORRETTA Luciano, PhD Student Tel : 32-2-650 5027
> Graphes et Optimisation Mathématique Fax : 32-2-650 5970
> Department of Computer Science e-mail : luciano.porretta at ulb.ac.be
> Université Libre de Bruxelles
> http://homepages.ulb.ac.be/~lporrett
> Building NO, Campus de la Plaine Boulevard du Triomphe CP 210 / 01
> Office: 2N3 202 B-1050 Brussels, Belgium
> -----------------------------------------------------------------------------------------
>
> On 15 Jul 2014, at 15:34, Ambros Gleixner <gleixner at zib.de> wrote:
>
>> Dear Luciano,
>>
>> you can use the lp/threads parameter,
>>
>> SCIP_CALL( SCIPsetIntParam(scip, "lp/threads", ???) );
>>
>> Best regards,
>> Ambros
>>
>>
>>
>> Am 15.07.2014 17:25, schrieb Luciano Porretta:
>>> Dear Scip users,
>>> I have implemented a MIP using cpp wrapper for column generation using
>>> CPLEX as LP-solver.
>>> I was wondering if it exists a way to interact with CPLEX options.
>>> I was looking inside the file: scip-3.1.0/src/lpi/lpi_cpx.c
>>> and the following functions:
>>> void* SCIPlpiGetSolverPointer(SCIP_LPI * lpi)
>>> SCIP_RETCODE SCIPlpiGetIntpar(SCIP_LPI * lpi, SCIP_LPPARAM type,int *
>>> ival )
>>> SCIP_RETCODE SCIPlpiSetIntpar(SCIP_LPI * lpi,SCIP_LPPARAM type, int
>>> ival )
>>> but actually i was wondering if exist a smarter and/or safer way to
>>> do it!
>>> Best regards,
>>> --
>>> -------------------------------------------------------------------------------------------------------------
>>> PORRETTA Luciano, PhD StudentTel : 32-2-650 5027
>>> Graphes et Optimisation MathématiqueFax : 32-2-650 5970
>>> Department of Computer Sciencee-mail : luciano.porretta at ulb.ac.be
>>> Université Libre de Bruxelleshttp://homepages.ulb.ac.be/~lporrett
>>> Building NO, Campus de la PlaineBoulevard du Triomphe CP 210 / 01
>>> Office: 2N3 202 B-1050 Brussels, Belgium
>>> --------------------------------------------------------------------------------------------------------------
>>>
>>>
>>>
>>> _______________________________________________
>>> Scip mailing list
>>> Scip at zib.de
>>> http://listserv.zib.de/mailman/listinfo/scip
>>>
>>
>> --
>> ____________________________________________________________
>> Ambros M. Gleixner
>> Zuse Institute Berlin - Matheon - Berlin Mathematical School
>> http://www.zib.de/gleixner
>> _______________________________________________
>> Scip mailing list
>> Scip at zib.de
>> http://listserv.zib.de/mailman/listinfo/scip
>
--
____________________________________________________________
Ambros M. Gleixner
Zuse Institute Berlin - Matheon - Berlin Mathematical School
http://www.zib.de/gleixner
More information about the Scip
mailing list