Thanks for the report. The tiMerge unit test case is too small to catch
performance regressions. I found the culprit and the performance should
now be back to its Amber14 levels (perhaps *slightly* faster).
The offending changes resulted in injecting a very expensive operation into
an inner loop (although the fact that it was expensive was very
non-obvious). Luckily, it was trivial to simply move that operation
outside the loops. Knowing that the regression was introduced between
Amber 14 and Amber 16 made it very easy to track down, so thanks Hannes for
that tidbit. The pull request on GitHub is here:
https://github.com/ParmEd/ParmEd/pull/851 and I will pull this into the
upcoming Amber 17 release (since tiMerge really is unusable as-is).
Thanks,
Jason
On Fri, Mar 17, 2017 at 5:38 AM, Hannes Loeffler <Hannes.Loeffler.stfc.ac.uk
> wrote:
> Hi Adrian and Dan,
>
> I just tried the A9 tutorial again and can confirm that it takes way to
> long (ParmEd from Amber16). But I have checked also with the old
> version from Amber 14 and it only took about 1 minute.
>
>
> Cheers,
> Hannes.
>
>
>
> On Thu, 16 Mar 2017 15:47:31 -0700
> Daniel Mermelstein <danmermelstein35.gmail.com> wrote:
>
> > Hi Adrian,
> >
> > I have one VERY ROUGH data point: I merged two copies of a 4300 atom
> > protein, and it took about a week. I don't know exactly how long
> > because I forgot it was still running...I also know an undergrad
> > working for me tried to run the sidechain timerge tutorial and gave
> > up after a couple days, in case that's useful
> >
> > Dan
> >
> > On Thu, Mar 16, 2017 at 7:48 AM, Adrian Roitberg <roitberg.ufl.edu>
> > wrote:
> >
> > > Dear all
> > >
> > > We are having trouble using the timerge function in parmed.
> > >
> > > In particular, if we try to follow the tutorials A9 for a change in
> > > side chain, the system hangs for a long time.
> > >
> > > We are testing some smaller systems, and it seems that the scaling
> > > with system size is really bad.
> > >
> > > Can someone give us some idea abou how long one might expect the
> > > timerge to take vs system size ?
> > >
> > > thanks
> > >
> > > adrian
> > >
> > >
> > >
> > > --
> > > Dr. Adrian E. Roitberg
> > > University of Florida Research Foundation Professor.
> > > Department of Chemistry
> > > University of Florida
> > > roitberg.ufl.edu
> > > 352-392-6972
> > >
> > >
> > > _______________________________________________
> > > AMBER mailing list
> > > AMBER.ambermd.org
> > > http://lists.ambermd.org/mailman/listinfo/amber
> > >
> >
> >
> >
>
>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
--
Jason M. Swails
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Sat Mar 18 2017 - 12:00:02 PDT
