Hi,

I have been trying out few sample examples in ULFM library. When I run one such 
example with OpenMPI on a cluster with infiniband interconnect using 32 nodes, 
512 processes

    mpirun –np 512 –am ft-enable-mpi ./simpleagree

I get this error,

[btl_openib_endpoint.c:694:mca_btl_openib_endpoint_connected] Error posting send

But when I execute the same program with fewer nodes (till 8 nodes, 128 
processes) it executes fine. I tried tuning the BTL parameters to tcp, sm, 
self, openib but I get the same error. Can someone let me know if I’m missing 
something here. Is this likely to be an issue with openmpi.

Thanks,
Murali

Reply via email to