[MOAB-dev] parallel i/o, not all vertices owned?

[Nico Schlömer]

> So in the beginning (without any ghosting), all shared vertices are owned
by either task 0 or 1

Aha, and since ghosting doesn't change ownership, after a ghosting step we
do have shared vertices in the highest-rank process (but only about half as
much as the other processes). Now makes all sense to me. Thanks!

--Nico




On Mon, Nov 2, 2015 at 5:51 AM Grindeanu, Iulian R. <iulian at mcs.anl.gov>
wrote:

> Sorry,
> I didn't see all these messages until now.
> these numbers refer only to vertices that are "shared" on  each local task
>
> there are vertices on a task that are not shared with anybody, so they are
> owned by the local process
>
> The example, again, refers only to shared entities
>
> after ghosting, the number of shared entities grows (as ghost layers are
> exchanged too)
>
> In moab, the convention is in general that the lower rank task owns the
> vertex if it is shared with other tasks.
>
> So in the beginning (without  any ghosting), all shared vertices are owned
> by either task 0 or 1
> ------------------------------
> *From:* moab-dev-bounces at mcs.anl.gov [moab-dev-bounces at mcs.anl.gov] on
> behalf of Nico Schlömer [nico.schloemer at gmail.com]
> *Sent:* Sunday, November 01, 2015 10:27 PM
> *To:* Timothy Tautges
> *Cc:* moab-dev at mcs.anl.gov
> *Subject:* Re: [MOAB-dev] parallel i/o, not all vertices owned?
>
> That's the thing: I only see 989+1078+679=2746 vertices out of a
> total 21218 owned. The rest is shared?
>
> Also, before ghosting, I see that process 3 doesn't own anything. Is that
> intentional?
>
> --Nico
>
> On Mon, Nov 2, 2015 at 5:16 AM Timothy Tautges <
> timothy.tautges at cd-adapco.com> wrote:
>
>> Many verts are owned and not shared.
>> On Nov 1, 2015 10:12 PM, "Nico Schlömer" <nico.schloemer at gmail.com>
>> wrote:
>>
>>> When using MOAB's Parallel Hello-World example on [1], a mesh with  over
>>> 20k vertices, I'm getting
>>> ```
>>> $ mpiexec -n 3 ./par-test
>>>  global rank:0 color:0 rank:0 of 3 processors
>>>  global rank:1 color:0 rank:1 of 3 processors
>>> Reading file ./2.h5m
>>>  with options:
>>> PARALLEL=READ_PART;PARTITION=PARALLEL_PARTITION;PARALLEL_RESOLVE_SHARED_ENTS
>>>  on 3 processors on 1 communicator(s)
>>>  Shared, owned entities on proc 0: 452 verts, 1162 edges, 711 faces, 0
>>> elements
>>>  Shared, owned entities on proc 1: 371 verts, 1037 edges, 666 faces, 0
>>> elements
>>>  Shared, owned entities on proc 2: 0 verts, 0 edges, 0 faces, 0 elements
>>>
>>>
>>>  After exchanging one ghost layer:
>>>  Shared, owned entities on proc 0: 989 verts, 1162 edges, 711 faces,
>>> 2301 elements
>>>  Shared, owned entities on proc 1: 1078 verts, 1037 edges, 666 faces,
>>> 3349 elements
>>>  Shared, owned entities on proc 2: 679 verts, 0 edges, 0 faces, 3182
>>> elements
>>> ```
>>> Only a few hundret owned vertices? That doesn't sum to to 20k. What's
>>> the issue here? Maybe I misunderstand the notion of ownage here.
>>>
>>> Cheers,
>>> Nico
>>>
>>> [1] http://chunk.io/f/c3ee1eebbcbd4734a368ffb287f1f37f
>>>
>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/moab-dev/attachments/20151102/7f5d31f8/attachment-0001.html>