[Beowulf] 1.2 us IB latency?

greg.lindahl at qlogic.com greg.lindahl at qlogic.com
Wed Apr 18 11:14:22 PDT 2007


On Wed, Apr 18, 2007 at 01:49:08PM -0400, Mark Hahn wrote:

> suggesting that your competitors are stupid or ignorant is _not_ good PR.

Yeah, this causes me to hold my tongue on a regular basis ;-)

I think Gil must have missed my email a while ago showing slides that
prove InfiniPath's cpu overhead is lower than Mellanox for smallish
MPI messages. It's well known that offload is often bad for small
messages; InfiniBand is not an exception to this rule.

-- greg






More information about the Beowulf mailing list