Viewing performance data with mmperfmon

To view the metrics that are associated with GPFS and the associated protocols, run the mmperfmon command with the query option. You can also use the mmperfmon command with the query option to detect performance issues and problems. You can collect metrics for all nodes or for a particular node.

  1. Problem: System slowing down

    Solution: Use the mmperfmon query compareNodes cpu_user command or the mmperfmon query compareNodes cpu_system command to compare CPU metrics for all the nodes in your system.

    • Check whether there is a node that has a significantly higher CPU utilization for the entire time period. If so, see whether this trend continues. You might need to investigate further on this node.
    • Check whether there is a node that has significantly lower CPU utilization over the entire period. If so, check whether that node has a health problem.
    • Use the mmperfmon query compareNodes protocolThroughput command to look at the throughput for each of the nodes for the different protocols.
      Note: The metrics of each individual protocol cannot always include exact I/O figures.
    • Use the mmperfmon query compareNodes protocolIORate command to look at the I/O performance for each of the nodes in your system.

  2. Problem: A particular node is causing problems

    Solution: Use the mmperfmon query usage command to show the CPU, memory, storage, and network usage.

  3. Problem: A particular protocol is causing problems

    Solution: Use the mmperfmon query command to investigate problems with your specific protocol. You can compare cross-node metrics by using the mmperfmon query compareNodes command.

    For example, the mmperfmon query compareNodes nfs_read_ops command compares the NFS read operations on all the nodes that are using NFS. By comparing the different NFS metrics, you can identify which node is causing the problems. The problem might either manifest itself as running with much higher values than the other nodes, or much lower (depending on the issue) when considered over several buckets of time.

  4. Problem: A particular protocol is causing problems on a particular node

    Solution: Use the mmperfmon query command on the particular node to look deeper into the protocol performance on that node.

    For example, if there is a problem with NFS:
    • Use the mmperfmon query nfsIOlatency command to get details of the nfsIOlatency.
    • Use the mmperfmon query nfsIOrate command to get details of the NFS I/O rate.
    • Use the mmperfmon query nfsThroughput command to get details of the NFS throughput.

For more information, see mmperfmon command .