Page 1 of 1

librdmacm - Unable to get RDMA device list

Posted: Tue Feb 23, 2016 1:52 am
by ffan
Hi,

I compiled CS 4.0.4 on a cluster machine. The mpi on the system is mvapich2. Everything was fine in build and installation. When run a case, it showed the following: (Please also see the attached.)
librdmacm: Warning: couldn't read ABI version.
librdmacm: Warning: assuming: 4
librdmacm: Fatal: unable to get RDMA device list

Despite the message says Fatal: unable to get RDMA device list, the run continued to completion. What do these librdmacm messages mean and how to fix them? Thank you.

- ffan

Re: librdmacm - Unable to get RDMA device list

Posted: Tue Feb 23, 2016 6:59 pm
by Yvan Fournier
Hello,

This seems to be a driver issue. If the computation runs normally, it is probably just a warning, though you might not be getting the maximum performance in some cases. In Code_Saturne, we mostly use MPI point-to-point and collective operations, and not one-sided operations, so RDMA issues are probably not a problem, but I am not knowledgeable enough in the low-level intricacies of various MPI libraries, and this is best checked with your system administrators or application support.

Regards,

Yvan