On Jan 16, 2013, at 7:41 AM, Jure Pečar <pega...@nerv.eu.org> wrote:
> > Hello, > > I have a large fortran code processing data (weather forecast). It runs ok > with smaller dataset, but on larger dataset I get some errors I've never seen > before: > > node061:05144] [[55141,0],11]->[[55141,0],0] mca_oob_tcp_msg_send_handler: > writev failed: Bad file descriptor (9) [sd = 9] > [node061:05144] [[55141,0],11] routed:binomial: Connection to lifeline > [[55141,0],0] lost This one means that a backend node lost its connection to mpirun. We use a TCP socket between the daemon on a node and mpirun to launch the processes and to detect if/when that node fails for some reason. > > and > > node084:7.0.Non-fatal temporary exhaustion of send tid dma descriptors > (elapsed=43.788s, source LID=0x49/context=11, count=1) (err=0) > > I'm using QLogic software version 7.1.0.0.58 (ofed 1.5.4.1, open-mpi 1.4.3). > > I'm starting this program with mpirun -mca btl openib,sm,self so I don't > really understand what tcp has to do in the first error message. > > Also I traced second error message to psm code, but it appears even if i add > -mca mtl ^psm to my mpirun arguments. Why? > > Any help appreciated. > > > -- > > Jure Pečar > http://jure.pecar.org > _______________________________________________ > users mailing list > us...@open-mpi.org > http://www.open-mpi.org/mailman/listinfo.cgi/users