interconnect latency, disected.

Greg Lindahl lindahl at
Sun Jun 29 15:25:33 EDT 2003

On Sun, Jun 29, 2003 at 12:37:45PM -0400, Mark Hahn wrote:

> 	- I haven't found any recent studies on where even 25 us of 
> 	GBE latency comes from.

store-and-forward swithes? It's entertaining to measure back-to-back
latency, but not very realistic.

> 	- are there current/active efforts to use something other than
> 	TCP for implementing the usual MPI primitives?  I'd love to see 
> 	something that used ethernet's broad/multicast support, for instance.

GAMMA and MVIA don't use TCP, but I don't think either has that. What
kind of savings do you expect, given the mismatch between unreliable
broadcast & multicast and MPI's semantics? I see a lot more enthusiasm
for such features from people who haven't looked at the details...


Beowulf mailing list, Beowulf at
To change your subscription (digest mode or unsubscribe) visit

More information about the Beowulf mailing list