[MOAB-dev] pcomm::exchange_tags unpredicted behaviour on large mesh

[Vijay S. Mahadevan]

Anton,

Can you try to replicate the issue on a smaller mesh? Perhaps there was an
issue with the way it was partitioned? What happens when you run on 32 or
lesser procs? What happens when you run on more than 64? Do you still see
the problem?

I'm trying to avoid downloading and replicating the issue on that large
mesh file until we understand the problem characteristics a little better.
So more info would help.

Vijay
On Jun 3, 2015 7:16 PM, "Kanaev A.A." <kanaev at ibrae.ac.ru> wrote:

> Hello, MOAB-dev,
>
> I have  10205296 hexa and  30736714 quads in my mesh
> I read the mesh on 64 processors and exchange 1 layer of ghost cells with
> additional 2d entities
> Then i exchange GLOBAL_ID tag data for quads
> After exchange it looks like all quads have received new GLOBAL_ID tag
> value, it doesn't appear on smaller meshes
> And there're two different quads with GLOBAL_ID = 30718357 on 41 proc
> after exchange (i think there're more doubles, that's just what i've
> spotted)
>
> Small test program attached
>
> The mesh is 3.9Gb, i hope that link works
> https://drive.google.com/file/d/0B6upEZeL2DFWXzhYZjEzNGlQem8/view
>
> Please check that on 64 processors
>
> Anton
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/moab-dev/attachments/20150603/550a9840/attachment.html>