Declustered array spares

For rebuilding from disk failures, the mmvdisk command configures spare space in each declustered array. This space is designated in whole pdisk units, but it is actually spread out across all of the pdisks of the declustered array.

As declustered arrays grow and shrink, the spare space is automatically adjusted to accommodate the new capacity and the existing vdisk sets.

For paired and shared recovery groups using twin-tailed disk enclosures, the number of spares depends on the number of pdisks:
  • One spare per enclosure when the total number of pdisks in the declustered array is greater than 24.
  • Two spares when the total number of pdisks is between 13 and 24.
  • One spare when the total number of pdisks is 12.
For scale-out recovery groups, the rules are more complicated since the declustered arrays can grow and shrink in units dependent on the number of pdisks each server contributes to each declustered array. In the default scale-out declustered array spares management:
  • The minimum number of equivalent spares in any scale-out declustered array is two. This permits IBM Storage Scale RAID to always be able to rebuild from two disk failures.
  • The maximum number of equivalent spares in any declustered array is two times the number of pdisks an individual server contributes to the declustered array. This permits IBM Storage Scale RAID to eventually be able to rebuild from two server failures.
  • As servers are added to scale-out recovery groups, the default number of spares automatically and gradually increases to approach the default maximum.

For scale-out recovery groups, the automatic default spares management can be overridden using the mmvdisk recoverygroup change command for all or some of the declustered arrays. Custom spares management can only be used to provide greater rebuild protection than default spares management; a scale-out recovery group cannot be customized to use fewer spares than would be configured by default management. Furthermore, customized spare settings are static and are not automatically adjusted when servers are added to or deleted from a scale-out recovery group. Depending on the vdisk sets defined in the scale-out recovery group, this can require manual adjustment of the custom spares when servers are added or deleted.

Custom spares for the declustered arrays of scale-out recovery groups can be specified in terms of:
  • Spare pdisks, meaning the exact number of equivalent pdisk spares to reserve; or
  • Spare nodes, meaning the number of pdisks contributed to the declustered array by a specified number of server nodes. This permits the declustered array to rebuild from the specified number of server failures.

The minimum number of custom spares in a scale-out declustered array can never be less than the default number of spares. The maximum number of custom spares in a scale-out declustered array is roughly equal to the number of pdisks supplied by one-third of the servers in the recovery group.

For more information on setting custom spares for scale-out recovery groups, see the mmvdisk recoverygroup man page.