FW: MPI cases and PVM

Donald Becker becker at scyld.com
Fri Mar 22 15:43:41 PST 2002


On Fri, 22 Mar 2002, Patrick Geoffray wrote:
> Daniel Kidger wrote:
>
> > However the biggest problem with layering on top of TCPIP is the issue of
> > latency.
>
> Right, and CPU load. TCP/IP is very hungry, having a machine on its knee
> just moving data is not very usefull for something else.

To amplify this message, TCP/IP overhead is increasing over time.

We have smarter NICs with hardware checksum and TCP packetization
offload, but new features such as IP chains, bandwidth throttling and
anti-denial-of-service have added greatly to the CPU work per packet and
cache impact of traversing the network stack.  Even if you don't use IP
chains, some of the overhead is still there.

TCP/IP is still an excellent protocol, and the overhead isn't bad for
1KB+ packets, but feature creep and security requirements have added
middle-aged "love handles" on it.  And IPv6 will just encourage an
eating binge.  [[ Oohhhh, bad middle-aged weight gain analogy. ]]

-- 
Donald Becker				becker at scyld.com
Scyld Computing Corporation		http://www.scyld.com
410 Severn Ave. Suite 210		Second Generation Beowulf Clusters
Annapolis MD 21403			410-990-9993




More information about the Beowulf mailing list