Error Reported If No NIC Is Specified During the Multi-Node OSU Collective Communication Test with Management and Compute LOMs and RDMA Driver Installed

Symptom

In the multi-node OSU collective communication test with management and compute LOMs and RDMA driver installed, if no NIC is specified, an error is reported, indicating that the traffic is transmitted through the management port.

[node35:1880003:0:1880003]       ud_ep.c:282  Fatal: UD endpoint 0xaaaae8a908c0 to :0: unhandled timeout error with local dev name:roceo1 remote dev gid:[subnet prefix:0 interface id:19004742ffff0000]

Possible Cause

The LOM driver identifies the management and compute network ports as RDMA ports. If HMPI processes select the management and compute network ports for communication, an error is reported because the management and compute networks are disconnected.

Recovery Procedure

When running the MPI application, specify only the compute network port as the available RDMA port.