IBM ServeRAID-4 Ultra160 Series, User Reference, Page: 115 / 208

v
Every logical drive that is shared by the two servers must have its cache policy
set to write-through mode to ensure that data integrity is maintained. Logical
drives that are not shared between the two servers can be configured for
write-back mode for improved performance.
v
Create only one logical drive for each array.
v
You must assign merge group numbers in the range 1–8 to each logical drive
that will be shared. Merge group numbers must be unique for each shared logical
drive in the cluster. You must assign merge group numbers 206 or 207 to the
non-shared logical drives.
v
If you are starting (booting) the operating system from a shared controller, define
the first logical drive as the startup drive and assign a merge group number for a
non-shared drive, for example, 206 for Server A.
v
The total number of logical drives per controller must be eight or less before or
after a failover. If you exceed this number, a failover will not complete.
v
The following logical drives will not failover:
– Logical drives currently undergoing Logical Drive Migration (LDM) operations
– RAID level-5E logical drives undergoing compression or decompression
– RAID level-5 logical drives in a critical state.
However, all other logical drives will failover if necessary.
v
If a failover occurs while a critical RAID level-1, RAID level-1E, RAID level-10 or
RAID level-1E0 logical drive is rebuilding to a spare disk, the Rebuild operation
automatically restarts.
v
The cluster support software will initiate a synchronization of all RAID level
logical drives except for the following:
– RAID level-0 logical drives
– Any critical or offline logical drives.
If a drive fails before this synchronization is complete, logical drive access is
blocked and is no longer accessible.
v
When a logical drive spans multiple SCSI channels and a failure within the drive
subsystem occurs that is unique to a channel (for example, a disconnected
cable), the entire physical array will be identified as failed even though access
from the surviving server can occur. Therefore, if you have small arrays, consider
not spanning multiple channels.
Chapter 12. Preparing to install or change a cluster solution
97