On Dec 8, 2006, at 5:41 AM, krishna chaitanya wrote:
I learnt from a reliable source that MPI uses the services provided by TCP/IP or infiniband. Suppose that there is a bottle-neck in the
Keep in mind that MPI is a specification -- a document. There are many software implementations of that specification, such as Open MPI. Open MPI can use a variety of network transports to effect the MPI semantics (e.g., the back end to MPI_SEND and MPI_RECV): TCP and IB are two, but Open MPI supports several others (Myrinet, shared memory, portals, ...). Additionally, there are other MPI implementations that support other networks that Open MPI does not currently support.
TCP/IP layer itself, how will optimization of MPI calls really help?
I'm not quite sure what you're asking here. Open MPI generally chooses the "best" network to transport most of its MPI traffic (both point-to-point and collectives). See this FAQ entry: http://www.open- mpi.org/faq/?category=tcp#tcp-auto-disable
Did that answer your question? -- Jeff Squyres Server Virtualization Business Unit Cisco Systems
