[Beowulf] Experience of using multiple network devices on a node in cluster

[Joe Landman]

Hi Tajendra:

   Latency will be a major issue if you channel bond.  If you are not 
doing that, you could (in theory) assign an MPI task per ethernet port. 
  There will likely be other issues you will worry about then, but this 
may be possible.

   Regardless, you will need a good switch to make any of this work, and 
you will need to make sure your nodes have different host names attached 
to the different nodes, and that mpi and the scheduler know about these 
, ...

   Rather than that, you might want to separate out your traffic into 
CnC (command and control), and messaging nets.  Have data transport 
(NFS, scp/rcp/...) over one net, and messages over the other.  In the 
gigE case this might not help all that much, apart from not swamping 
connections that are already being pounded by IO requests.

Joe

Tajendra Singh wrote:
> Greetings!!
> 
> Many of the vendors are providing two GigE ports by default on systems which
> are being used as nodes in beowulf types of clusters.
> 
> Is there anybody who tried connecting both the ports to the switch? I know
> LA-MPI ( http://public.lanl.gov/lampi/) talks about utilizing multiple
> network devices in real time, but did not get chance to experiment with it.
> I would like to hear your opinion whether it makes sense, looking at the
> lower costs per port for GigE switches?
> 
> Thanks in advance,
> 
> TV Singh  
> HPC Applications, ATS/UCLA
> 
> 
> _______________________________________________
> Beowulf mailing list, Beowulf at beowulf.org
> To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf

-- 
Joseph Landman, Ph.D
Founder and CEO
Scalable Informatics LLC,
email: landman at scalableinformatics.com
web  : http://www.scalableinformatics.com
phone: +1 734 786 8423
fax  : +1 734 786 8452
cell : +1 734 612 4615