A kernel level message passing service

Greg Lindahl glindahl at hpti.com
Fri Sep 1 13:29:09 PDT 2000


> (I wonder sometimes
> if the heavy machinery of TCP really makes sense in the context
> of private, switched networks.  does anyone have a count of dropped
> frames inside a normal cluster?)

No, but I can give you lists of machines which were unusable because they
assumed such errors were very infrequent, and didn't deal with them. ASCI
Blue Mountain is one example. They had to add a rather expensive and
annoying reliability layer to correct the problem.

Issues like that are why Myricom's GM driver provides reliable, in-order
delivery, even though it hurts their small-message latency.

If you want to use unreliable ethernet user-level networking, there's
already one out there.

-- g





More information about the Beowulf mailing list