[Beowulf] interconnect and compiler ?

[Greg Lindahl]

On Fri, Jan 30, 2009 at 10:28:29PM +0100, Håkon Bugge wrote:

> Greg, we (and your former colleagues at PathScale) have exchanged  
> opinions on RDMA vs. Message Passing. Based on SPEC MPI2007, you will  
> find that an RDMA based DDR interconnect using Platform MPI performs  
> better than a Message Passing stack using DDR.

Well, like any experiment in this space, this isn't quite
head-to-head.  It could easily be the case that Platform MPI has some
better collectives or other implemtation details which provide a
performance boost outside of the RDMA vs. message passing issue. I
thought you had a PSM implementation of Platform MPI? In which case,
perhaps you could compare it against the QLogic MPI on the same
hardware, to see what these other effects might be.

It would also be interesting to compare Platform MPI against
MVAPICH on the same hardware -- do you have those numbers?

-- greg