On Wed, 22 Jan 2003, Jeremy Enos wrote:
> Each node doesn't have an entry for every other node... only the server,
> which is doing NAT by default. So this isn't going to ever affect MPI
> or node to node performance at all. As for the server, it's probably
> not a big enough problem to justify the complication yet... the entries
> you pasted look to me like they must be from a server.
Yes, they are from the server -- sorry, I didn't clarify that.
I'm not concerned about MPI communication (etc.) -- but what about other
traffic? For example: getting my data to/from the OSCAR cluster (i.e.,
the head node), or if the nodes NFS mount machines that are routed through
the head node? (recall that we had a lengthy dinner conversation at
Dagwood's sub shop at IU last week where Neil was expounding on major IO
requirements for each job) And so on.
Even if there isn't a huge performance degration, it would be good to get
this feature on the request list, anyway, because one rule per node really
doesn't seem to be the Right Thing. That's why we have masks, after all.
--
{+} Jeff Squyres
{+} [EMAIL PROTECTED]
{+} http://www.lam-mpi.org/
-------------------------------------------------------
This SF.net email is sponsored by: Scholarships for Techies!
Can't afford IT training? All 2003 ictp students receive scholarships.
Get hands-on training in Microsoft, Cisco, Sun, Linux/UNIX, and more.
www.ictp.com/training/sourceforge.asp
_______________________________________________
Oscar-devel mailing list
[EMAIL PROTECTED]
https://lists.sourceforge.net/lists/listinfo/oscar-devel
