<div class="gmail_quote">On Sun, Nov 27, 2011 at 3:28 PM, Matthew Knepley <span dir="ltr"><<a href="mailto:knepley@gmail.com">knepley@gmail.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<div class="im">On Sun, Nov 27, 2011 at 2:25 PM, Sean Farley <span dir="ltr"><<a href="mailto:sean@mcs.anl.gov" target="_blank">sean@mcs.anl.gov</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">
<div class="gmail_quote"><div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">So there would not be the problem with empty domains? I could see it happening for this many processes.</blockquote>
<div><br></div></div><div>I'm saying that he's using the older parmetis (3.1) and SuperLU_DIST (2.5) (and also Jed was right, this looks like 3.2p1), which as far I remember still have the empty domain problem.</div>
</div>
</blockquote></div><br></div>Okay, executive summary:<div><br></div><div>There is a problem with the interaction of SuperLU and ParMetis which is fixed in petsc-dev, using the latest</div><div>releases of both packages. Consider upgrading.</div>
</blockquote><div><br>Sorry to dig up an old thread (I seem to have a tendency of doing that here!), but this problem exists in petsc-3.2-p6 with MUMPS and Metis as well. Looking at the differences between the dev version of mump.c and the release version, I don't see anything that would have obviously fixed this. Could you give me a hint about what was changed? Is it as simple as using the updated Metis package (in dev)? How difficult would it be to backport the fix to the released version?<br>
<br>John<br><br></div></div>