Hi Jeff, Given the situation that 16 nodes have eth0,eth1 and eth2 interface, I tried to run data transfer within themselves using mpirun, but without specifying "btl_tcp_if_include". I got only 15% increase in uni-directional data transfer rate when using 3 links. But if I run two such processes : one using eth0+eth1 and other using eth2 only, I get expected 50% increase in tranfer rate. Any clue?
Regards, Jayanta On Wed, Sep 2, 2009 at 1:41 PM, Jeff Squyres <jsquy...@cisco.com> wrote: > If you don't use btl_tcp_if_include, Open MPI should use all available > ethernet devices, and *should* (although I haven't tested this recently) > only use devices that are routable to specific peers. Specifically, if > you're on a node with eth0-3, it should use all of them to connect to > another peer that has eth0-3, but only use eth0-1 to connect to a peer that > only has those 2 devices. (all of the above assume that all your eth0's are > on one subnet, all your eth1's are on another subnet, ...etc.) > > Does that work for you? > > > > On Aug 25, 2009, at 7:14 PM, Jayanta Roy wrote: > > Hi, >> >> I am using Openmpi (version 1.2.2) for MPI data transfer using >> non-blocking MPI calls like MPI_Isend, MPI_Irecv etc. I am using "--mca >> btl_tcp_if_include eth0,eth1" to use both the eth link for data transfer >> within 48 nodes. Now I have added eth2 and eth3 links on the 32 compute >> nodes. My aim is to share the high speed data within 32 compute nodes >> through eth2 and eth3. But I can't include this as part of "mca" as the rest >> of 16 nodes do not have these additional interfaces. In MPI/Openmp can one >> specify explicit routing table within a set of nodes? Such that I can edit >> /etc/host for additional hostname with these new interfaces and add these >> hosts in the mpi hostfile. >> >> Regards, >> Jayanta _______________________________________________ >> users mailing list >> us...@open-mpi.org >> http://www.open-mpi.org/mailman/listinfo.cgi/users >> > > > -- > Jeff Squyres > jsquy...@cisco.com > > > _______________________________________________ > users mailing list > us...@open-mpi.org > http://www.open-mpi.org/mailman/listinfo.cgi/users >
