[MPICH] Intercommunicator limits

[Rajeev Thakur]

> 1)    Is there a limit to the number of intercommunicators you can have
and, if so, are we going over that limit if we have 24 B nodes

 

I don't think there is a limit. You should be able to run 24.

 

> 2)       Would a single machine (our cluster is down at the moment) be
able to handle all the MPI communications for this system (for testing
purposes only, until our cluster is back up)?

 

You should be able to test it on one machine although it will crawl.

 

Rajeev


  _____  

From: owner-mpich-discuss at mcs.anl.gov
[mailto:owner-mpich-discuss at mcs.anl.gov] On Behalf Of Yoder, Jeffrey - SSD
Sent: Friday, March 31, 2006 8:56 AM
To: MPICH Discussion Group
Subject: [MPICH] Intercommunicator limits



We have an architecture where we have a single process (A), with multiple
send threads, that is parsing a data file and parcelling it out, using a
load-balancing scheme, to n different processing nodes (B1 - Bn).  These B
nodes each also have a dedicated receive thread associated to the A node by
means of an intercommunicator.  A similar setup is used on the backend where
the B nodes have dedicated send threads all tied via intercommunicators to
receive threads in node C where the processed data is all collected and put
into our output format.

 

Our questions are:

1)    Is there a limit to the number of intercommunicators you can have and,
if so, are we going over that limit if we have 24 B nodes

2)       Would a single machine (our cluster is down at the moment) be able
to handle all the MPI communications for this system (for testing purposes
only, until our cluster is back up)?



-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/mpich-discuss/attachments/20060331/2e61d28c/attachment.htm>