Notes common to OSs

The following notes are common to OSs:

• Do not activate the same volume group concurrently from multiple servers.

• When a path error exists, the period of time required to return a response to the application that made an I/O request to HDLM depends on the timeout values specified for the Fibre Channel switch and HBA.

  Therefore, if an error occurs in all the paths for a single LU, this response time will be the total period of time required for the corresponding Fibre Channel switches and HBAs to detect a timeout for the I/O requested for each path.

  For details on the timeout values for Fibre Channel switches and HBAs, see the appropriate manual for the device. The following figure gives an overview of the period required to return a response from the time an application requests an I/O to the time the application receives the response.

  Figure. Overview of the period required to respond to an application's I/O request

  As shown in the preceding diagram, when an HDLM device has two paths (SCSI devices), the maximum period of time required to respond to the application's I/O request is n1 + n2 seconds; where n1 indicates the timeout value specified for the path that uses SCSI device A, and n2 indicates the timeout value specified for the path that uses SCSI device B.

• When using HDLM in a cluster environment, a node in the cluster may fail over before the path completes failover. To avoid this, when you set up the failover timeout value for the node by using cluster software, make sure that you specify a period longer than the response time that is calculated as described in the preceding note.

• If a path error occurs while creating a file system, or formatting, or executing fsck, the operation may not finish. In such a case, perform the operation again after restoring the path error.

• The name of the HDLM device file will not be displayed on the Hardware Browser of Red Hat Enterprise Linux.

• The Linux functionality that adds LABEL= or UUID= to a SCSI device is not supported in HDLM. Do not use this functionality. With HDLM, access can constantly be made to the same LU if the name of the HDLM device file is the same.

• If all the following conditions are satisfied, even a path in the normal status might become Offline(E) or Online(E):

  • A storage system that has a SATA drive is being used.

  • There is a heavy I/O load on the storage system.

  • Path health checking is set to ON.

  If this problem happens, reduce the I/O load, and then return the path status to Online by performing an online operation.

• Although the following messages are output when you load the alert driver and the filter driver respectively, HDLM operations are not affected.

  sddlmadrv: unsupported module, tainting kernel.
  sddlmadrv: module license 'Proprietary: Contact your HDLM vendor or the maintenance company for any bugs' taints kernel.

• An HDLM device supports a maximum of 15 partitions.

  You cannot use an HDLM device with more than 15 partitions created in an LU.

• In an environment where HDLM is installed (a multi-path configuration), processing takes longer than in an environment where HDLM is not installed (a single-path configuration) because of the following reasons. When building a system, perform a load test to verify that there are no performance problems. If there is a problem, consider revising the number of connection paths or using an LVM environment.

  • When an HDLM device receives an open command from a higher-level system, open processing for the device is performed as many times as the number of connection paths.

  • In the open processing for the device, information about the connection destination devices is obtained. For this reason, if the open or close processing is performed multiple times, it can take time.

• SCSI devices corresponding to an HDLM device can be removed dynamically even while the HDLM device is in use. Before doing so, perform the following operations:

  1. Unmount any devices mounted on the HDLM device.

  2. Execute the dlmcfgmgr -o command to exclude the HDLM device from management.

• When a device is dynamically configured by using the udev function, the messages shown below might be output to the console or syslog. However, HDLM operations are not affected.

  Output example:

  scsi_id[a01]: scsi_id: unable to access parent device of '/block/sddlma01'

  For a01 in the above examples, any alphanumeric character can be output.