Storage system requirements for FlashCopy, volume mirroring, and thin-provisionedvolumes

Application performance on a local clustered system can be affected by the use of FlashCopy, volume mirroring, and thin-provisionedvolumes for storage systems.

The FlashCopy, volume mirroring, and thin-provisionedvolume functions can all have a negative impact on system performance. The impact depends on the type of I/O taking place, and is estimated by using a weighting factor from Performance impact estimates for FlashCopy, volume mirroring, and thin-provisionedvolumes.

A FlashCopy mapping effectively adds a number of loaded volumes to the storage pool. The effect of mirrored and thin-provisionedvolumes is also estimated in Performance impact estimates for FlashCopy, volume mirroring, and thin-provisionedvolumes. The estimates assume that thin-provisionedvolumes are running at approximately 80% capacity of a fully allocated volume, and that mirrored volumes read from one copy and write to all copies.

Table 1. Performance impact estimates for FlashCopy, volume mirroring, and thin-provisionedvolumes
Type of I/O (to volume)	Impact on I/O weighting	FlashCopy weighting	Volume mirroring weighting	Thin-provisioned weighting
None or minimal	Insignificant	0	0	0
Read only	Insignificant	0	0	0.25 * Sv
Sequential read and write	Up to 2 x I/O	2 * F	C−V	0.25 * Sc
Random read and write	Up to 15 x I/O	14 * F	C−V	0.25 * Sc
Random write	Up to 50 x I/O	49 * F	C−V	0.25 * Sc
Notes: In a storage pool with two FlashCopy mappings and random read/write to those volumes, the weighting factor is 14 * 2 = 28. In a storage pool with 10 copies, five of which are primary copies of a volume, a weighting factor of 10−5 = 5 applies. If the copies are thin-provisioned, an extra weighting factor of 0.25 * 10 = 2.5 applies. Key: C Number of volume copies in this storage pool. V Number of volumes with their primary copy in this storage pool. F Number of FlashCopy mappings that are affecting volumes that have copies in this storage pool. Sv Number of thin-provisionedvolume copies in this storage pool that are the primary copy of a volume. Sc Number of thin-provisionedvolume copies in this storage pool.

To calculate the average I/O rate per volume, use the following equation:

I/O rate = (I/O capacity) / ( V + weighting factor for FlashCopy + 
weighting factor for volume mirroring + weighting factor for thin-provisioned)

For example, consider 20 volumes with an I/O capacity of 5250, a FlashCopy weighting of 28, a mirroring weighting of 5, and a thin-provisioned weighting of 0.25. The I/O rate per volume is 5250 / (20 + 28 + 5 + 2.5) = 94.6. This estimate is an average I/O rate per volume; for example, half of the volumes might be running at 200 I/O operations per second (IOPs), and the other half might be running at 20 IOPs. However, this rate would not overload the system because the average load is 94.6.

If the average I/O rate to the volumes in the example exceeds 94.6, the system would be overloaded. As approximate guidelines, a heavy I/O rate is 200, a medium I/O rate is 80, and a low I/O rate is 10.

With volume mirroring, a single volume can have multiple copies in different storage pools. The I/O rate for such a volume is the minimum I/O rate that is calculated from each of its storage pool.

If system storage is overloaded, you can migrate some of the volumes to storage pools with available capacity.

Note: Flash drives are exempt from these calculations, except for overall node throughput, which increases substantially for each additional Flash drive in the node.