FW: MPI cases and PVM

[Daniel Kidger]

Robert G. Brown [mailto:rgb at phy.duke.edu] wrote:
>One area where MPI might hold a small advantage is in low-level network
>device support, specifically Myrinet.  I think MPI has native Myrinet
>drivers and can avoid TCP altogether.  I don't really know if PVM does
>also at this point (although somebody on the list that uses Myrinet
probably does:-).  

All of the highest performance interconnects, (including both our own and
Myricom), do MPI and Shmem directly using a native comms layer. 

Both Quadrics and Myricom can also layer TCPIP over their interconnects, and
get close to the theroetic peak bandwidth of the cabling that way.

However the biggest problem with layering on top of TCPIP is the issue of
latency. In running an application over a large number of processors, IMHO
99% of the time it is the latency that ultimately limits scalability.

I do not know of a cluster interconnect that supports PVM any other way than
on top of TCPIP.





Yours,
Daniel.

--------------------------------------------------------------
Dr. Dan Kidger, Quadrics Ltd.      daniel.kidger at quadrics.com
One Bridewell St., Bristol, BS1 2AA, UK         0117 915 5505
----------------------- www.quadrics.com --------------------