[Scip] Scip 3.1.0 is much more slower than 3.0.2

[Gregor Hendel]

Dear Vladimir

In addition to what Marc wrote, I suggest that you check SCIP's 
changelog <http://scip.zib.de/doc-3.1.0/html/CHANGELOG.php> which 
contains a section about changed parameters between the versions.

I see that on your particular instance the number of solving nodes has 
increased by about a factor of 15, which should focus your attentions on 
changes w.r.t. branching.
Between versions, we decreased the parameter 
branching/relpscost/maxreliable from 8.0 to 5.0 which might be 
disadvantageous for your problem. You may try to reset the parameter to 
8.0 or even a higher value and check if this improves the performance.

Hope this helps,
Gregor

Am 06.06.2014 19:09, schrieb Marc Pfetsch:
>
>
> Hi Vladimir,
>
> it is likely that you experienced what is called performance 
> variability: Small changes in the implementation of the (LP-)solver 
> (or input) can lead to a completely different solution process. This 
> cannot really be avoided.
>
> More information can be found in answers to the following email to 
> this list:
>
> http://listserv.zib.de/pipermail/scip/2013-November/001734.html
>
> If the newer version is consistently slower on a very large testset, 
> however, the reason possibly is more structural.
>
> Best
>
> Marc
>
>
> On 06.06.2014 16:34, Vladimir VV wrote:
>> Dear SCIP team,
>> I was surprised when found than last version of Scip works more slowly
>> than previous 3.0.2 did.
>> For example, the same MILP problem of the following dimensions (from
>> scipampl log):
>> "... 134 variables (114 bin, 0 int, 0 impl, 20 cont) and 1159 
>> constraints"
>> has been solved by SCIP 3.0.2 in ~170 sec.
>> "Solving Time (sec) : 169.58
>> Solving Nodes      : 210199
>> Primal Bound       : +8.38888888888988e+00 (51 solutions)"
>> and by SCIP 3.1.0 in ~900 sec.
>> "Solving Time (sec) : 891.05
>> Solving Nodes      : 3333047
>> Primal Bound       : +8.38888888888929e+00 (41 solutions)
>> Dual Bound         : +8.38888888888929e+00"
>>
>> Both solvers had run with the default parameters settings at the same 
>> host.
>> But looking at solvers' outputs I see that 3.1.0 processes the search
>> tree completely differently than 3.0.2 does.
>>
>> The same effect I see with many other problems... What is the reason 
>> of that?
>>
> _______________________________________________
> Scip mailing list
> Scip at zib.de
> http://listserv.zib.de/mailman/listinfo/scip

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.zib.de/pipermail/scip/attachments/20140610/5de1d0b4/attachment.html>