Follow these guidelines when setting up your disk configuration and the associated md.tab file.
It is advisable to maintain identical md.tab files on each node in the cluster to ease administration.
A multihost disk and all the partitions found on that disk can be included in no more than one diskset.
All metadevices used by data services must be fully mirrored. Two-way mirrors are recommended, but three-way mirrors are acceptable.
No components of a submirror for a given mirror should be found on the same controller as any other component in any other submirror used to define that mirror.
If more than two disk strings are used, each diskset must include disks from at least three separate controllers. If only two disk strings are used, each diskset must include disks from the two controllers and mediators will be configured. See the Sun Cluster 2.2 System Administration Guide for more information about using dual-string mediators.
Hot spares are recommended, but not required. If hot spares are used, configure them so that the activation of any hot spare drive will not result in components of a submirror for a given metamirror sharing the same controller with any other component in any other submirror used to define that given metamirror.
If you are using Solaris UFS logging, you only need to set up mirrored metadevices in md.tab files, transdevices are not necessary.
If your are using Solstice DiskSuite logging, create multihost file systems on trans metadevices only. Both the logging and master device components of each trans metadevice must be mirrored.
If you are using Solstice DiskSuite logging, do not share spindles between logging and master device components of the same trans metadevice, unless the devices are striped across multiple drives. Otherwise, system performance will be degraded.
Each diskset has a small administrative file system associated with it. This file system is not NFS shared. It is used for data service-specific state or configuration information. The administrative file system should be named the same as the logical host that is the default master for the diskset. This strategy is necessary to enable start up of DBMS fault monitors.