[Beowulf] SC|05: will 10Ge beat IB?

Greg Lindahl lindahl at pathscale.com
Tue Nov 22 17:30:04 PST 2005


On Tue, Nov 22, 2005 at 10:59:47AM -0500, Gary Green wrote:

> The second issue is the lack of a broadly adopted RDMA implementation.  It
> appears that there is movement forward in this arena as well with the move
> to adopt the verbs interface from the OpenIB group.

People with MPI clusters want them to run fast. I see a lot of people
assuming "RDMA must be good when you want to go fast", but in this
case it's rather unproven. If RDMA isn't fast, nobody's going to care
about it in an MPI context... perhaps for I/O.

OpenIB does offer an interesting entre' for non-IB stuff, that's
certainly true.

I suppose you do touch on the RDMA performance issue in that you talk
about performance separately:

> There is also the question as to whether GigE will be able to demonstrate
> the ultra low latencies seen in high performance interconnects such as
> Quadrics, InfiniBand and Myrinet.  In the end, there will most likely remain
> a market for high performance interconnects for high end applications.

Amen!

-- greg




More information about the Beowulf mailing list