fast blackplanes (was Re: Experience with SMC 8624T L2 switches and nics)
math at velocet.ca
Wed Feb 26 23:54:31 PST 2003
On Wed, Feb 26, 2003 at 02:44:03PM -0800, Dale Harris's all...
> On Wed, Feb 26, 2003 at 01:38:36PM -0800, Greg Lindahl elucidated:
> > > Are there any implementations using the chip that you would advise?
> > I'm afraid not -- my interest in the part was theoretical, so once I
> > knew it existed, I was done. It would be really if someone could do a
> > survey of the current switch marketplace; back in the 100 mbit days,
> > someone would do one about every 6 months and post the results.
> Yeah, that'd be nice if someone had those resources.
> Off hand, it appears now to be cheaper per port to go with Myrinet with a
> high number of nodes (128+ node cluster), than it is with a big GigE
> capable switch.
If you need interconnects between all nodes. Which means you application
must scale to that many nodes, otherwise I guess you might have it
for convenience (when running n jobs that scale to 128/n nodes only), but
thats steep for convenience.
> Dale Harris
> rodmur at maybe.org
> Beowulf mailing list, Beowulf at beowulf.org
> To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
Ken Chase, math at velocet.ca * Velocet Communications Inc. * Toronto, CANADA
More information about the Beowulf