On Wed, Dec 22, 2010 at 6:01 AM, Thomas Witkowski <span dir="ltr"><<a href="mailto:thomas.witkowski@tu-dresden.de">thomas.witkowski@tu-dresden.de</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;">
So, I found the problem related to empty partitions. It is not possible to weight vertices (i.e. elements of a mesh) in such a way that one weight is much higher than the other ones. For more details see<br>
<br>
<a href="http://glaros.dtc.umn.edu/flyspray/task/11" target="_blank">http://glaros.dtc.umn.edu/flyspray/task/11</a><br>
<br>
Its a pity that ParMetis makes is very hard to find this kind of errors.<br>
<br>
The open question for me is about the non continuous partitions. Is it a normal behavior of ParMetis to create partitions that are not continous?</blockquote><div><br></div><div>Yes, this is normal.</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;"><font color="#888888"><br>
Thomas</font><div><div></div><div class="h5"><br>
<br>
Thomas Witkowski wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
Okay, in my computations, I have empty partitions on some ranks and definitely not<br>
minimal boundary sizes. So may be I generate a wrong input. But if this is the case, I<br>
wonder why the resulting mesh partitioning is quite good. If I neglect the problem of<br>
empty partitions, the redistributed mesh leads to a very good load balancing. Is there<br>
any meaningful way to debug the problem? Is there something link a "verbose mode" in<br>
ParMetis that says me whats happen on the input data? Otherwise I have to print all the<br>
input data to the screen and check it by hand. Although I have a quite small example with<br>
128 overall coarse mesh elements on 8 ranks, this is not big fun :)<br>
<br>
Thomas<br>
<br>
@Matthew: By mistake I've answered your mail directly to you and not to the mailing list, therefore I sent it now here again<br>
<br>
Matthew Knepley wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
On Tue, Dec 21, 2010 at 5:49 AM, Thomas Witkowski <<a href="mailto:thomas.witkowski@tu-dresden.de" target="_blank">thomas.witkowski@tu-dresden.de</a> <mailto:<a href="mailto:thomas.witkowski@tu-dresden.de" target="_blank">thomas.witkowski@tu-dresden.de</a>>> wrote:<br>
<br>
Hi,<br>
<br>
I have a not directly PETSc related question, but I hope to get<br>
some answer from the community here. In my FEM code, I make use of<br>
ParMETIS to partition the mesh. I make direct use of this library<br>
and not of PETSc's ParMETIS integration. The initial partition is<br>
always fine, but I use the ParMETIS_V3_AdaptiveRepart function for<br>
repartition the mesh due to local mesh adaption. In most cases,<br>
the result is fine, but there are two points, where I have trouble<br>
with:<br>
<br>
1) Sometimes ParMETIS generates empty partitions, i.e., a<br>
processor has zero mesh elements. This is something my code cannot<br>
handle. Is this a bug or a feature? If it is a feature, is there<br>
any possiblity to disable it?<br>
<br>
<br>
ParMetis has a balance constraint if you weight vertices. This will enforce equal size partitions.<br>
<br>
2) In most cases the specific partitions are not connected. If I<br>
put all data to ParMETIS in a correct way, is this okay? My code<br>
can handle it, but is slows down the computation due to larger<br>
interior boundaries and therefore to more communications.<br>
<br>
<br>
ParMetis minimizes the overall boundary size, so I do not understand how you could see this slowdown.<br>
<br>
Matt<br>
<br>
Does anyone of you know an answer to these question? Is there a<br>
debug mode in ParMETIS, where I can see which data is set to its<br>
function calls?<br>
<br>
Regards,<br>
<br>
Thomas<br>
<br>
<br>
<br>
<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>
</blockquote>
<br>
<br>
<br>
</blockquote>
<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>