[MPICH] Doing collectives a better way in not-so-small nodes

[Joachim Worringen]

Rajeev Thakur schrieb:
> Sylvain,
>         Your approach sounds good. We have been meaning to add
> topology-aware collectives to MPICH2 for a while now, but haven't gotten
> around to it. We hope to add something next year.
> 
> Rajeev 
> 
>> -----Original Message-----
>> From: owner-mpich-discuss at mcs.anl.gov 
>> [mailto:owner-mpich-discuss at mcs.anl.gov] On Behalf Of Sylvain Jeaugey
>> Sent: Tuesday, November 28, 2006 3:12 AM
>> To: mpich-discuss at mcs.anl.gov
>> Subject: [MPICH] Doing collectives a better way in not-so-small nodes
>>
>> Hi all,
>>
>> When using not-so-small nodes, which means with more than 4 
>> cores, most
>> collective operations often suffer a lot from intra-node contention. 
>> Therefore, the best performance may be obtained using 
>> different algorithm 
>> for intra-node and inter-node.

Sylvain, this paper gives a good overview of what can be (and has been) 
  done in this area in case you want to work in this area. With the 
upcoming multi-cores, this will become much more important than today.

@InProceedings{,
author =        {Jesper-Larsson Träff and Hubert Ritzdorf},
title =         {Collective Operations in NEC's High-performance MPI 
Libraries},
booktitle =     {IPDPS, Rhodes, Greece, 25 - 29.04.2006},
year =          {2006},
publisher =     {IEEE},
keywords =      {MPI, collective operations, broadcast, reduction, 
all-to-all communication, MPI/SX, MPI/ES     },
}

  Joachim

-- 
Joachim Worringen, Software Architect, Dolphin Interconnect Solutions
phone ++49/(0)228/324 08 17 - http://www.dolphinics.com