You really have to give us the log summary output. None of the relevant<br>numbers are in your summary.<br><br> Thanks,<br><br> Matt<br><br><div><span class="gmail_quote">On 2/9/07, <b class="gmail_sendername">Shi Jin</b>
<<a href="mailto:jinzishuai@yahoo.com">jinzishuai@yahoo.com</a>> wrote:</span><blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;">Dear Barry,
<br><br>Thank you.<br>I actually have done the staging already.<br>I summarized the timing of the runs in google online<br>spreadsheets. I have two runs.<br>1. with 400,000 finite elements:<br><a href="http://spreadsheets.google.com/pub?key=pZHoqlL60quZeDZlucTjEIA">
http://spreadsheets.google.com/pub?key=pZHoqlL60quZeDZlucTjEIA</a><br>2. with 1,600,000 finite elements:<br><a href="http://spreadsheets.google.com/pub?key=pZHoqlL60quZcCVLAqmzqQQ">http://spreadsheets.google.com/pub?key=pZHoqlL60quZcCVLAqmzqQQ
</a><br><br>If you can take a look at them and give me some<br>advice, I will be deeply grateful.<br><br>Shi<br>--- Barry Smith <<a href="mailto:bsmith@mcs.anl.gov">bsmith@mcs.anl.gov</a>> wrote:<br><br>><br>> NO, NO, don't spend time stripping your code!
<br>> Unproductive<br>><br>> See the manul pages for PetscLogStageRegister(),<br>> PetscLogStagePush() and<br>> PetscLogStagePop(). All you need to do is maintain a<br>> seperate stage for each<br>> of your KSPSolves; in your case you'll create 3
<br>> stages.<br>><br>> Barry<br>><br>> On Fri, 9 Feb 2007, Shi Jin wrote:<br>><br>> > Thank you.<br>> > But my code has 10 calls to KSPSolve of three<br>> > different linear systems at each time update.
<br>> Should I<br>> > strip it down to a single KSPSolve so that it is<br>> > easier to analysis? I might have the code dump the<br>> > Matrix and vector and write another code to read<br>> them<br>
> > into and call KSPSolve. I don't know whether this<br>> is<br>> > worth doing or should I just send in the messy<br>> log<br>> > file of the whole run.<br>> > Thanks for any advice.<br>
> ><br>> > Shi<br>> ><br>> > --- Barry Smith <<a href="mailto:bsmith@mcs.anl.gov">bsmith@mcs.anl.gov</a>> wrote:<br>> ><br>> > ><br>> > > Shi,<br>> > ><br>> > > There is never a better test problem then
<br>> your<br>> > > actual problem.<br>> > > Send the results from running on 1, 4, and 8<br>> > > processes with the options<br>> > > -log_summary -ksp_view (use the optimized<br>> version of
<br>> > > PETSc (running<br>> > > config/configure.py --with-debugging=0))<br>> > ><br>> > > Barry<br>> > ><br>> > ><br>> > > On Fri, 9 Feb 2007, Shi Jin wrote:
<br>> > ><br>> > > > Hi there,<br>> > > ><br>> > > > I am tuning our 3D FEM CFD code written with<br>> > > PETSc.<br>> > > > The code doesn't scale very well. For example,
<br>> > > with 8<br>> > > > processes on a linux cluster, the speedup we<br>> > > achieve<br>> > > > with a fairly large problem size(million of<br>> > > elements)<br>> > > > is only 3 to 4 using the Congugate gradient
<br>> > > solver. We<br>> > > > can achieve a speed up of a 6.5 using a GMRes<br>> > > solver<br>> > > > but the wall clock time of a GMRes is longer<br>> than<br>> > > a CG
<br>> > > > solver which indicates that CG is the faster<br>> > > solver<br>> > > > and it scales not as good as GMRes. Is this<br>> > > generally<br>> > > > true?<br>
> > > ><br>> > > > I then went to the examples and find a 2D<br>> example<br>> > > of<br>> > > > KSPSolve (ex2.c). I let the code ran with a<br>> > > 1000x1000<br>> > > > mesh and get a linear scaling of the CG solver
<br>> and<br>> > > a<br>> > > > super linear scaling of the GMRes. These are<br>> both<br>> > > much<br>> > > > better than our code. However, I think the 2D<br>> > > nature
<br>> > > > of the sample problem might help the scaling<br>> of<br>> > > the<br>> > > > code. So I would like to try some 3D example<br>> using<br>> > > the<br>> > > > KSPSolve. Unfortunately, I couldn't find such
<br>> an<br>> > > > example either in the<br>> > > src/ksp/ksp/examples/tutorials<br>> > > > directory or by google search. There are a<br>> couple<br>> > > of<br>> > > > 3D examples in the
<br>> src/ksp/ksp/examples/tutorials<br>> > > but<br>> > > > they are about the SNES not KSPSolve. If<br>> anyone<br>> > > can<br>> > > > provide me with such an example, I would
<br>> really<br>> > > > appreciate it.<br>> > > > Thanks a lot.<br>> > > ><br>> > > > Shi<br>> > > ><br>> > > ><br>> > > ><br>> > > >
<br>> > ><br>> ><br>><br>____________________________________________________________________________________<br>> > > > Finding fabulous fares is fun.<br>> > > > Let Yahoo! FareChase search your favorite
<br>> travel<br>> > > sites to find flight and hotel bargains.<br>> > > ><br>> <a href="http://farechase.yahoo.com/promo-generic-14795097">http://farechase.yahoo.com/promo-generic-14795097</a><br>
> > > ><br>> > > ><br>> > ><br>> > ><br>> ><br>> ><br>> ><br>> ><br>> ><br>><br>____________________________________________________________________________________
<br>> > 8:00? 8:25? 8:40? Find a flick in no time<br>> > with the Yahoo! Search movie showtime shortcut.<br>> > <a href="http://tools.search.yahoo.com/shortcuts/#news">http://tools.search.yahoo.com/shortcuts/#news
</a><br>> ><br>> ><br>><br>><br><br><br><br><br>____________________________________________________________________________________<br>Looking for earth-friendly autos?<br>Browse Top Cars by "Green Rating" at Yahoo! Autos' Green Center.
<br><a href="http://autos.yahoo.com/green_center/">http://autos.yahoo.com/green_center/</a><br><br></blockquote></div><br><br clear="all"><br>-- <br>One trouble is that despite this system, anyone who reads journals widely
<br>and critically is forced to realize that there are scarcely any bars to eventual<br>publication. There seems to be no study too fragmented, no hypothesis too<br>trivial, no literature citation too biased or too egotistical, no design too
<br>warped, no methodology too bungled, no presentation of results too<br>inaccurate, too obscure, and too contradictory, no analysis too self-serving,<br>no argument too circular, no conclusions too trifling or too unjustified, and
<br>no grammar and syntax too offensive for a paper to end up in print. -- Drummond Rennie