A kernel level message passing service

Walter B. Ligon III walt at parl.ces.clemson.edu
Fri Sep 1 14:23:13 PDT 2000


> for cluster programming.  note that this topic is different from VIA-type
> user-level hardware access.  and if you want a low-overhead message
> passing scheme that's still protected in the conventional sense, 
> why not use good old UDP or even IP datagrams.  (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?)

Actually, we have a little project going to do that, more or less.  We're
actually pitching IP altoghter, but staying with the existing Linux network
protocol framework to implement a lightweight, reliable messaging service
that continues to work with the existing sockets interface.  It requires
very little change to user code to use it.  The question is - how much
performance savings is due to the protocol, and how much is due to the
interface?  I don't claim to know the answer, so we figured we'd try it out
and see what happens.


Dr. Walter B. Ligon III
Associate Professor
ECE Department
Clemson University

More information about the Beowulf mailing list