Manually installing the performance monitoring tool

The performance monitoring tool is automatically installed by the installation toolkit. You can also install the performance monitoring tool manually for the releases that are not supported by the installation toolkit.

For more information, see Understanding the installation toolkit options. For more information about the performance monitoring tool, see Configuring the performance monitoring tool.

A single collector can easily support up to 150 sensor nodes. The collector can be any node on the system. All sensors report to this node. Select any node in the system to be the collector node. For information on configuring the sensors and collectors, see Configuring the performance monitoring tool. You can manually install the tool by doing the following steps:
  1. Download the installation images and install the performance monitoring packages.

    For information on the default directories in which these packages are extracted, see Location of extracted packages.

    Performance monitoring packages

    gpfs.gss.pmsensors-version_release.os.target_arch.file_format
    gpfs.gss.pmcollector-version_release.os.target_arch.file_format
    Then, use your operating system’s native package management mechanism to install the packages.

    For example:

    • To install 5.0.*-0 sensors on a Red Hat 7 x86_64 node, use the following command:
      rpm -ivh gpfs.gss.pmsensors-5.0.*-0.el7.x86_64.rpm
    • To install 5.0.*-0 sensors on a SLES 12 x86_64 node, use the following command:
      rpm -ivh gpfs.gss.pmsensors-5.0.*.sles12.x86_64.rpm
    • To install 5.0.*-0 sensors on an Ubuntu amd64 node, use the following command:
      dpkg -i gpfs.gss.pmsensors_5.0.*-0.U*amd64.deb
    • To install a 5.0.*-0 collector on a Red Hat 7 PPC64LE node, use the following command:
      rpm -ivh gpfs.gss.pmcollector-5.0.*-0.el7.ppc64le.rpm
  2. To configure performance monitoring after the packages have been installed on all the selected nodes, use the following commands:
    mmperfmon config generate --collectors collectorNode1[,collectorNode2,...]
    mmchnode --perfmon –N sensorNode1[,sensorNode2...]
  3. To enable performance monitoring on a protocol node for NFS, SMB, or Object, follow the given steps.
    • For NFS:
      1. Install the pm-ganesha package as follows:
        • RHEL or SLES: rpm -ivh gpfs.pm-ganesha_version-release.os.target_arch.rpm
        • Ubuntu: dpkg -i gpfs.pm-ganesha_version-release.deb
      2. Ensure that the /opt/IBM/zimon/defaults/ZIMonSensors_nfs.cfg sensor is created as shown:
        
        sensors={
                name = "NFSIO"
                period = 10
                type = "Generic"
                restrict = "cesNodes"
        }
      3. Run the following command to add the sensor:
        mmperfmon config add --sensors /opt/IBM/zimon/defaults/ZIMonSensors_nfs.cfg
      4. Run the following code to ensure that the newly added sensor is made visible to the system:
        mmhealth node show nfs --refresh -N cesNodes
    • For SMB:
      1. Run the following command to add the sensor:
        mmperfmon config add --sensors /opt/IBM/zimon/defaults/ZIMonSensors_smb.cfg
        Note: No additional sensor package is needed for the SMB sensors.
      2. Run the following code to ensure that the newly added sensor is made visible to the system:
        mmhealth node show smb --refresh -N cesNodes
    • For Object:
      1. Install the pmswift package as follows:
        • RHEL: rpm -ivh pmswift-version-release.noarch.rpm
        • Ubuntu: dpkg -i pmswift_version-release.deb

          Where version is equal to or greater than 4.2 and release is equal to or greater than 0.

          The installation of the pmswift RPM also copies SWIFT related sensors configuration files, namely, SwiftAccount.cfg, SwiftContainer.cfg, SwiftObject.cfg, and SwiftProxy.cfg to the performance monitoring tool’s installation directory, /opt/IBM/zimon/. The pmswift package converts the operational metrics for Object into a form that is usable by the performance monitoring tool.

      2. Edit the Object configuration files for all Object servers that reside in the cluster configuration repository (CCR), using the following command:
        /usr/local/pmswift/bin/pmswift-config-swift set
        CCR then propagates the modified configuration files to /etc/swift/ directory on all the protocol nodes within the cluster. The modified configuration files are:
        • account - *.conf
        • container - *.conf
        • object - *.conf
        • proxy - *.conf
      3. Use the /usr/local/pmswift/bin/pmswift-config-zimon set command to edit the sensors configuration information stored in the CCR. This adds the SWIFT related following sensors entries:
        {
                # SwiftAccount operational metrics
                name = "SwiftAccount"
                period = 1
                type = "generic"
        },
        {
                # SwiftContainer operational metrics
                name = "SwiftContainer"
                period = 1
                type = "generic"
        },
        {
                # SwiftObject operational metrics
                name = "SwiftObject"
                period = 1
                type = "generic"
        },
        {
                # SwiftProxy operational metrics
                name = "SwiftProxy"
                period = 1
                type = "generic"
        },
        These entries are then automatically propagated to the ZIMonSensors.cfg file in /opt/IBM/zimon on all the nodes in the cluster.
      4. Start the pmswiftd.service by using the following command:
        systemctl start pmswiftd.service
      5. Start or restart the pmsensors.service by using the following command:
        systemctl start|restart pmsensors.service
    For more information on how to manually upgrade pmswift, see the Manually upgrading pmswift topic.
  4. If the protocol sensors are enabled on a GPFS-only node, you see an error message stating that the sensors are unavailable. However, the other sensors continue to run.
  5. Start the sensors on each node using the systemctl start pmsensors.service command.
  6. On the collector nodes, start the collector, using the systemctl start pmcollector.service command.
  7. To ensure that sensors and collectors are restarted after the node reboots, you can enable them using the following commands:
    Sensors
    To enable sensors, use the systemctl enable pmsensors.service command.
    To disable sensors, use the systemctl disable pmsensors.service command.
    Collector
    To enable the collector, use the systemctl enable pmcollector.service command.
    To disable the collector, use the systemctl disable pmcollector.service command.
The collector node starts gathering all the requested metrics.
Important: Once the packages are installed, pmsensor and pmcollector services are activated automatically.
Note: Although you can enable sensors on every node in a system, with the increase in number of nodes, the metric collection work for the collector also increases. It is recommended to ensure that collection of metrics does not increase above 1000000 metrics per second.
By default, the installation toolkit enables sensors on each protocol node (CES node) but not on the other GPFS nodes (non-CES nodes).

Metrics can be retrieved from any node in the system by using the mmperfmon query command. For more information, see mmperfmon command.

For more information about the performance monitoring tool, see Configuring the performance monitoring tool.