Chapter 5. Maintenance and serviceability 285Example 5-23 Stopping LoadLevelerxdsh compute -v -l loadl llrctl stopxdsh service -v -l loadl llctl stopStopping GPFS and unmounting the file systemAfter LoadLeveler is stopped, GPFS also is stopped. It is important to ensure that allapplications that need to access files within GPFS are stopped before performing this step. Asingle command is run on any storage node to complete this step. For this example, we use astorage node called f01st01 (for frame one storage node one), as shown in Example 5-24.Example 5-24 Stopping GPFSxdsh f01st01 -v mmshutdown -aOptional: Capturing the HFI lInk statusBefore the compute LPARs are shut down, it might be useful to get a state of the HFI linkstatus. This status is useful if there are any HFI errors before shutting down so that the errorsare understood when the cluster is restarted. This restart is done by listing the connectionstate for the BPAs and FSPs and listing the CEC link status.Verify that CNM successfully contacted all BPAs and FSPs by issuing the command that isshown in Example 5-25.Example 5-25 Checking the status of all the BPAs and FSPslsnwcomponentsExample 5-26 must match the number of CEC drawers that are in the cluster.Example 5-26 Matching the number of the CEC drawerlsnwloc | grep -v EXCLUDED | wc -lIf the number is incorrect, check for any issues that cause a CEC drawer to be excluded byCNM, as shown in Example 5-27.Example 5-27 Checking issues with CEC$ xdsh f01st01 -v mmshutdown -a$ lsnwloc | grep EXCLUDEDShutting down compute nodesNow that LoadLeveler and GPFS are stopped, the next step is to shut down the computenodes, as shown in Example 5-28. The compute nodes are shut down first because othernodes within the cluster do not depend on the compute nodes.Example 5-28 Shutting down the compute nodesxdsh compute -v shutdown -h nowImportant: This command shuts down GPFS everywhere it is running in the cluster. Afterthe command completes, GPFS is down and no longer available.