IBM Power Systems 775 Manual page 96

Typically, a JBOD array is divided into two recovery groups that are controlled by different
primary GPFS Native RAID servers. If the primary server of a recovery group fails, control
automatically switches over to its backup server. Within a typical JBOD, the primary server for
a recovery group is the backup server for the other recovery group.
Figure 1-60 illustrates the ring configuration where GPFS Native RAID servers and storage
JBODs alternate around a loop. A particular GPFS Native RAID server is connected to two
adjacent storage JBODs and vice versa. The ratio of GPFS Native RAID server to storage
JBODs is thus one-to-one. Load on servers increases by 50% when a server fails.
Figure 1-60 GPFS Native RAID server and recovery groups in a ring configuration
Declustered arrays
A declustered array is a subset of the physical disks (pdisks) in a recovery group across
which data, redundancy information, and spare space are declustered. The number of disks
in a declustered array is determined by the RAID code-width of the VDisks that are housed in
the declustered array. One or more declustered arrays can exist per recovery group.
Figure 1-61 on page 83 illustrates a storage JBOD with two recovery groups, each with four
declustered arrays.
A declustered array can hold one or more VDisks. After redundancy codes are associated
with VDisks, a declustered array simultaneously contains Reed-Solomon and replicated
VDisks.
If the storage JBOD supports multiple disks that are physically mounted together on
removable carriers, removal of a carrier temporarily disables access to all of the disks in the
carrier. Thus, pdisks on the same carrier must not be in the same declustered array, as VDisk
redundancy protection is weakened upon carrier removal.
Declustered arrays are normally created at recovery group creation time but new arrays are
created or existing arrays are grown by adding pdisks later.
82
IBM Power Systems 775 for AIX and Linux HPC Solution