[Symphony] Symphony solver question

[Ted Ralphs]

Such performance variability can certainly occur. Changing the order of the
data leads to ties being broken in different ways and other perturbations
to the algorithm that can substantially effect its evolution. Small
differences in the initial phases can lead to big differences later,
especially for very difficult instances like the ones you are apparently
facing. Section 5 in this paper discusses the phenomena:

http://coral.ie.lehigh.edu/~ted/files/papers/miplib5.pdf

See especially the figures on page 14 of he above paper illustrating that
the amount of variability can be very different for different instances.
Note that variability is significantly higher in the case of parallel
computation, which introduces additional sources of variability. SYMPHONY
runs in parallel by default, so you may be seeing the effects of that,
which can arise even without changing the order of the data. See, e.g.,
Section 4.3 here:

*http://coral.ie.lehigh.edu/~ted/files/papers/Million11.pdf
<http://coral.ie.lehigh.edu/~ted/files/papers/Million11.pdf>*

There are various ways to control this variability, but making a solver run
deterministically typically degrades performance overall and we have chosen
not o implement this in SYMPHONY. There is a discussion of this in Section
3.2.7 here:

http://coral.ie.lehigh.edu/~ted/files/papers/ParallelMILPSurvey16.pdf

Hope this helps!

Cheers,

Ted

On Thu, Nov 30, 2017 at 8:59 PM, Bill Magan <bill.magan at gmail.com> wrote:

> We are running what should be a fairly basic optimization for 13 million
> records with 24 constraints on our matrix.  Is there any reason why the
> time to reach the first feasible solution would vary by several hours
> (sometimes runs in 2 hours other times more than 8 hours) based on the
> order of the data ?
>
> Thank you,
> Bill
>
> Sent from my iPhone
> _______________________________________________
> Symphony mailing list
> Symphony at list.coin-or.org
> https://list.coin-or.org/mailman/listinfo/symphony
>



-- 
Dr. Ted Ralphs
Professor and Interim Department Chair
Industrial and Systems Engineering
Lehigh University
(610) 628-1280
ted 'at' lehigh 'dot' edu
coral.ie.lehigh.edu/~ted
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/symphony/attachments/20171202/c2960115/attachment.html>