Optimal storage pool configuration guidelines for storage systems

A storage pool provides the pool of storage from which volumes are created. You must ensure that the MDisks that make up each tier of the storage pool have the same performance and reliability characteristics.

Notes:
  1. The performance of a storage pool is generally governed by the slowest MDisk in the storage pool.
  2. The reliability of a storage pool is generally governed by the weakest MDisk in the storage pool.
  3. If a single MDisk in a pool fails, access to the entire pool is lost.
Use the following guidelines when you group similar disks:
  • Group MDisks with equal performance in a single tier of a pool.
  • Group similar arrays in a single tier. For example, configure all 6 + P RAID-5 arrays in one tier of a pool.
  • Group MDisks from the same type of storage system in a single tier of a pool.
  • Group MDisks that use the same type of underlying physical disk in a single tier of a pool. For example, group MDisks by Fibre Channel or SATA.
  • Do not use single disks. Single disks do not provide redundancy. Failure of a single disk results in total data loss of the storage pool to which it is assigned.
Important: When grouping internal RAID MDisks into pools using similar technology disks, the arrays need to use a similar slow_write_priority setting to prevent creating a single array that cannot avoid writing to a slow component. Failure to do this would result in an immediate critical performance bottleneck if a single drive in the pool became slow in the redundancy mode array. A simple exception to this rule is when using the IBM Easy Tier function, the flash drive arrays are set to redundancy mode (as the write response time technique is too coarse for typical flash drive latencies) and the spinning disk arrays are set to latency mode.

Scenario: Similar disks are not grouped together

Under one scenario, you could have two externalstorage systems that are attached behind your system. One device is an IBM TotalStorageEnterprise Storage Server (ESS), which contains ten 6 + P RAID-5 arrays and MDisks 0 through 9. The other device is an IBM System StorageDS5000, which contains a single RAID-1 array, MDisk10, one single JBOD, MDisk11, and a large 15 + P RAID-5 array, MDisk12.

If you assigned MDisks 0 through 9 and MDisk11 into a single storage pool, and the JBOD MDisk11 fails, you lose access to all of the IBM ESS arrays, even though they are online. The performance is limited to the performance of the JBOD in the IBM DS5000storage system, therefore slowing down the IBM ESS arrays.

To fix this problem, you can create three pools. The first pool must contain the IBM ESS arrays, MDisks 0 through 9, the second pool must contain the RAID 1 array, and the third pool must contain the large RAID 5 array.