Checking for fabric configuration and functional problems

Use this information to learn how to check for fabric configuration and functional problems.

To check for fabric configuration and functional problems, perform the following procedure.

On the fabric management server run the all_analysis fast fabric health check command. For details, see Health checking. To diagnose symptoms reported by health check see Table 3.

Note: The health check is most effective for checking for configuration problems if a baseline health check has been performed and is stored in the /var/opt/iba/analysis/baseline directory on the fabric management server. Otherwise changes in configuration cannot be sensed.

If there is no baseline health check for comparison, you need to perform the same type of configuration checks that were done during installation. For details, see Installing and configuring the InfiniBand switch. For the host-based subnet managers, also use the Installing the fabric management server topic. You need to check that the following configuration parameters match the installation plan. A reference or setting for IBM® System p® and IBM Power Systems™ HPC Clusters is provided for each parameter that you check.

Table 1. Health check parameters
Parameter	Reference or setting
GID prefix	The GID prefix must be different for each subnet. For details, see Planning for global identifier prefixes.
LMC	Must be 2 for IBM HPC Clusters.
Maximum transfer unit (MTU)	For details, see Planning for maximum transfer units (MTUs). This parameter is the fabric MTU and not the MTU in the stack, which can be a much greater number.
Cabling plan	See the vendor Switch Users Guide and Planning and Installation Guide
Balanced Topology	It is typically best to ensure that you have distributed the HCA ports from the servers in a consistent manner across subnets. For example, all corresponding ports on HCAs within servers must connect to the same subnet; similar to, all port 1 on HCA 1 must connect to subnet 1, and all port 2 on HCA 1 must connect to port 2.
Full bandwidth topology?	Did you choose to implement a full-bandwidth topology by using the vendor recommendations found in the vendor Switch Users Guide and Planning and Installation Guide?