Log groups

Every recovery group contains one or more log groups. A log group is a collection of vdisks that share the same log home vdisk (the RAID transaction log).

Paired recovery groups have only one log group. A paired recovery group is therefore identical with its single log group. Workload is balanced in a recovery group pair by making each server primary for one of the paired recovery groups and backup for the other.

Scale-out and shared recovery groups have two log groups per server, plus a special root log group that is only used for recovery group metadata and event logs. Workload is balanced among the servers of scale-out and shared recovery groups by making each server responsible for the vdisk NSDs of two log groups.

Vdisk sets balance workload across recovery groups by placing one member vdisk NSD in each log group of each recovery group in which the vdisk set is defined.

The combination of log groups and vdisk sets automatically ensures several IBM Storage Scale RAID best practices:

Each vdisk set in a file system is balanced across the servers for each recovery group in the vdisk set.
Each log group of a recovery group contains one member of each vdisk set in the recovery group.
Each log group of a recovery group therefore represents the same quantity of NSD I/O responsibility.
Each log group of a recovery group therefore uses the same capacity from each declustered array in the recovery group.
In the case of scale-out or shared recovery groups, each server is assigned two log groups.
In the case of a recovery group pair, each server is primary for one log group (one paired recovery group).
Therefore, each server is nominally responsible for an equal quantity of I/O load and disk capacity across the recovery group or recovery group pair.

The balance of log group assignments can be shown with the mmvdisk server list or mmvdisk recoverygroup list commands.

For paired recovery groups, the balance is best shown using mmvdisk server list with the node class for the recovery group pair since the node class spans both recovery groups of the pair:


# mmvdisk server list --node-class ESS01

 node
number  server                            active   remarks
------  --------------------------------  -------  -------
     1  ess01io1.gpfs.net                 yes      serving ESS01L
     2  ess01io2.gpfs.net                 yes      serving ESS01R

For a scale-out or shared recovery group, the mmvdisk recoverygroup list command with the --server option can also be used to show all the log groups, since the log groups are contained within a single recovery group:


# mmvdisk recoverygroup list --recovery-group RG01 --server

 node
number  server                            active   remarks
------  --------------------------------  -------  -------
     1  server01                          yes      serving RG01: LG004, LG008
     2  server02                          yes      serving RG01: LG002, LG006
     3  server03                          yes      serving RG01: LG003, LG007
     4  server04                          yes      serving RG01: root, LG001, LG005