Those error messages are from OpenMPI. Make sure your application runs correctly outside of SGE first, and then set the same resource limits in the shell and test your application again before even trying to submit your application to SGE. The OpenMPI library needs internal memory to handle send/rec buffers, so I assume that when OpenMPI applications are executed under a low enough resource limit the applications would break.)
If your application is not happy with the resource limits or another things in the environment, then running the application under SGE with the same limits or environment will also fail. (In the end, SGE does not know anything about the application unless you tell it about the specifics.) Rayson On Sun, Nov 27, 2011 at 1:16 AM, mahbube rustaee <[email protected]> wrote: > Hi, > I run a job with this requirements: > #!/bin/bash > #$ -S /bin/bash > #$ -N Det2 > #$ -cwd > #$ -j y > #$ -pe mpifillamd 50 > #$ -l mem_free=70M,h_vmem=100M > mpirun -n $NSLOTS mem.comp > > job start running and output is created , but a tranmission of "r" state to > "dr" state did. > sometimes output is ok, but sometimes at the end of output some messages > appear: > [amd-8-9.local:30368] [[8294,0],3]-[[8294,0],0] mca_oob_tcp_msg_recv: readv > failed: Connection reset by peer (104) > [amd-8-9.local:30368] [[8294,0],3] routed:binomial: Connection to lifeline > [[8294,0],0] lost > [amd-8-7.local:29707] [[8294,0],1]-[[8294,0],0] mca_oob_tcp_msg_recv: readv > failed: Connection reset by peer (104) > [amd-8-7.local:29707] [[8294,0],1] routed:binomial: Connection to lifeline > [[8294,0],0] lost > > and these errors are not same and depend on mem_free and h_vmem values in > script! > > what does it mean? > > Thx > > > _______________________________________________ > users mailing list > [email protected] > https://gridengine.org/mailman/listinfo/users > > _______________________________________________ users mailing list [email protected] https://gridengine.org/mailman/listinfo/users
