On Thu, Aug 12, 2010 at 2:01 PM, John Hernlund <span dir="ltr"><<a href="mailto:hernlund@gmail.com">hernlund@gmail.com</a>></span> wrote:<br><div class="gmail_quote"><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex;">
Hi Shijie,<br>
<br>
I think this is a good thing to do, since,<br>
<br>
1) there is a lot of discussion about the speed of various codes, but<br>
no comprehensive quantitative comparisons have been performed to my<br>
knowledge,<br>
<br>
2) solver efficiency is an important benchmark of progress in mantle<br>
convection-related computing and code development and is one of our<br>
most important computational issues in general, and,<br>
<br>
3) if we can sort out which solution approaches work best (e.g.,<br>
treatment of pressure, various multigrid schemes, etc.) then we can<br>
perhaps understand which direction future development might want to go.<br>
<br>
It is not always simple to compare speed for different codes running<br>
on different architectures and different CPUs (unless you know of a<br>
good way of measuring the number of floating point operations, FPOs,<br>
which is the more robust measure), and running these cases will take a<br>
little time, but perhaps for the above 3 reasons everyone with a code<br>
can run some of these comparisons.<br></blockquote><div><br></div><div>4) PAPI is a nice and easy way to measure the floating point operations. It</div><div> works particularly well on some architectures (BG, Linux), and you only</div>
<div> need to run it once to get your count.</div><div><br></div><div>5) All the time measurements are necessary for a benchmark, but just as</div><div> crucial is a complete specification of the stopping criteria for the iterative</div>
<div> solver. Without this, it is impossible to compare the efficiency of different</div><div> solvers and preconditioners.</div><div><br></div><div> Thanks,</div><div><br></div><div> Matt</div><div> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex;">
Paul Tackley (cc) or somebody in his group might be able to run these<br>
comparisons relatively quickly using his code, and this could be a<br>
good place to begin.<br>
<br>
Cheers!<br>
<font color="#888888">John<br>
</font><div><div></div><div class="h5"><br>
On Aug 12, 2010, at 10:19 AM, Shijie Zhong wrote:<br>
<br>
><br>
> I apologize in advance if this topic is not of interest to you.<br>
><br>
> At the GLADE meeting, there were some good discussions on the need<br>
> for more<br>
> benchmark, particularly regarding timing and speed. For CitcomS, our<br>
> 2008 G^3<br>
> paper provides quite extensive benchmark calculations for Stokes'<br>
> flow,<br>
> isochemical and thermochemical convection with large viscosity<br>
> contrast (up to<br>
> 1e7). In Table 1 of this paper, we also include some information on<br>
> the speed<br>
> and CPU time for a typical calculation, and scaling up to 3000 cores.<br>
><br>
> Needless to say, more benchmarks and comparisons with other codes<br>
> will be<br>
> helpful. If any of you are interested in doing this, I would be more<br>
> than happy to<br>
> participate or take the lead. I can provide all those solutions<br>
> published in the<br>
> paper including analytical solutions for Stokes' flow.<br>
><br>
><br>
><br>
> Shijie Zhong<br>
> Department of Physics<br>
> University of Colorado at Boulder<br>
> Boulder, CO 80309<br>
> Tel: 303-735-5095; Fax: 303-492-7935<br>
> Web: <a href="http://anquetil.colorado.edu/szhong" target="_blank">http://anquetil.colorado.edu/szhong</a><br>
<br>
_______________________________________________<br>
CIG-MC mailing list<br>
<a href="mailto:CIG-MC@geodynamics.org">CIG-MC@geodynamics.org</a><br>
<a href="http://geodynamics.org/cgi-bin/mailman/listinfo/cig-mc" target="_blank">http://geodynamics.org/cgi-bin/mailman/listinfo/cig-mc</a><br>
</div></div></blockquote></div><br><br clear="all"><br>-- <br>What most experimenters take for granted before they begin their experiments is infinitely more interesting than any results to which their experiments lead.<br>
-- Norbert Wiener<br>