[Beowulf] Really efficient MPIs??

[Scott Atchley]

On Nov 28, 2007, at 8:49 AM, Charlie Peck wrote:

> On Nov 28, 2007, at 8:04 AM, Jeffrey B. Layton wrote:
>
>> Unless you are using a gigabit ethernet, Open-MPI is noticeably  
>> less efficient that LAM-MPI over that fabric.
>
> I suspect at some point in the future gige will catch-up but for  
> now my (limited) understanding is that the Open-MPI folks are  
> focusing their time on higher bandwidth/lower latency fabrics than  
> gige.
>
> charlie


At SC07 MPICH2 BoF, I gave a brief talk about MPICH2-MX. In addition  
to showing results of it running over MX-10G, I had a few slides  
showing performance using MPICH2-MX over Open-MX on Intel e1000  
drivers (80003ES2LAN NICs). IMB Pingpong latency is ~10 us for small  
messages and throughput for large messages is near line rate.

Open-MX provides the same API as MX, runs on _any_ Ethernet driver/ 
NIC, should work with most MX software (MPICH-MX, MPICH2-MX, PVFS2,  
etc.) and is open source. It has not reached a release tarball yet,  
but we expect to see one soon.

Scott