[MOAB-dev] deadlock in MBErrorCode MBParallelComm::exchange_ghost_cells

[Lukasz Kaczmarczyk]

It looks strange, without reason I set #define MAX_SHARING_PROCS 200  
and I get

Entering exchange_ghost_cells with num_layers = 0
Posting Irecv from 0 for ghost entities.
Posting Irecv from 1 for ghost entities.
Posting Irecv from 2 for ghost entities.
Posting Irecv from 6 for ghost entities.
Posting Irecv from 12 for ghost entities.
Posting Irecv from 13 for ghost entities.
Posting Irecv from 14 for ghost entities.
Posting Irecv from 20 for ghost entities.
Posting Irecv from 23 for ghost entities.
Posting Irecv from 24 for ghost entities.
Posting Irecv from 25 for ghost entities.
Posting Irecv from 12 for 2nd message.
Posting Irecv from 13 for 2nd message.
Posting Irecv from 14 for 2nd message.
Posting Irecv from 23 for 2nd message.
Posting Irecv from 0 for 2nd message.
Posting Irecv from 1 for 2nd message.
Posting Irecv from 24 for 2nd message.
Posting Irecv from 25 for 2nd message.
Posting Irecv from 25 for 2nd message.
Posting Irecv from 27 for 2nd message.
Posting Irecv from 28 for 2nd message.
Posting Irecv from 7 for 2nd message.
Posting Irecv from 23 for 2nd message.
Posting Irecv from 24 for 2nd message.
Posting Irecv from 30 for 2nd message.
Posting Irecv from 12 for 2nd message.
Posting Irecv from 6 for 2nd message.
Exiting exchange_ghost_cells

And deadlock at the end.


On 7 Oct 2009, at 01:13, Lukasz Kaczmarczyk wrote:

> This is output with: #define DEBUG_COMM
>
> Entering exchange_ghost_cells with num_layers = 0
> Posting Irecv from 0 for ghost entities.
> Posting Irecv from 1 for ghost entities.
> Posting Irecv from 2 for ghost entities.
> Posting Irecv from 6 for ghost entities.
> Posting Irecv from 12 for ghost entities.
> Posting Irecv from 13 for ghost entities.
> Posting Irecv from 14 for ghost entities.
> Posting Irecv from 20 for ghost entities.
> Posting Irecv from 23 for ghost entities.
> Posting Irecv from 24 for ghost entities.
> Posting Irecv from 25 for ghost entities.
> Posting Irecv from 12 for 2nd message.
> Posting Irecv from 13 for 2nd message.
> Posting Irecv from 14 for 2nd message.
> Posting Irecv from 0 for 2nd message.
> Posting Irecv from 1 for 2nd message.
> Posting Irecv from 23 for 2nd message.
> Posting Irecv from 24 for 2nd message.
> Posting Irecv from 25 for 2nd message.
>
>
>
> On 7 Oct 2009, at 01:01, Lukasz Kaczmarczyk wrote:
>
>> Hello
>>
>> I run a mesh partitioned for 32 processors and I spot deadlock in  
>> MBParallelComm::exchange_ghost_cells. This functions is called by  
>> MBParallelComm::resolve_shared_ents.
>>
>> I can give link to mesh if you need it for tests. For 8 processor I  
>> get correct results.
>>
>> Regards,
>> Lukasz
>>
>>
>>
>
>

Lukasz Kaczmarczyk
Lecturer
Department of Civil Engineering,
University of Glasgow,
GLASGOW, G12 8LT
Tel: +44 141 3305348
email: likask at civil.gla.ac.uk
web: http://www.civil.gla.ac.uk/~kaczmarczyk/
web: http://code.google.com/p/yaffems/