EMC allows more than 128 hypers per disk; finding a swap that improves service time by 10 percent or more on these systems is very rare.
Since using a percentage of improvement as swap goodness criteria is not applicable anymore, Optimizer had to adapt a different method (M0) that deals better with the “many hypers contribute smaller chunks” problem. M0 is defined as “the best you could possibly get” from a swap. Each disk in the system is assigned an M0 designation, which is defined as the minimum service time a disk can get by replacing one of its hypers by a null hyper (a hyper that does no I/Os to the disk). When analyzing a swap, Optimizer checks how close the new-modeled service time is to the disk’s M0; the closer to M0, the better the swap is. Usually, two RAID groups are affected by a single swap; the busier disk’s service time is expected to go down while the other disk’s service time is expected to go up. In addition to the M0, the Optimizer also ensures that the new maximum of service times is less than the old one.
The swap procedure relies on the Symmetrix TimeFinder technology and uses DRVs as temporary mirrors. The following section describes the four swap steps when dealing with mirrored devices swaps. In the case of three mirror devices or RAID 6 devices the swap procedure does not require a DRV.
Mirrored device swap steps
Hypervolumes are swapped using a four-step process. In order to swap two hypervolumes, the hypervolumes must be the same size, the same emulation, and, based on the architecture, on the same system bus or in the same power zone in the Symmetrix DMX™ architecture in versions prior to 5772. At least two DRVs must be configured per swap. With Symmetrix Optimizer 5.1.1, up to eight simultaneous swaps can happen at one time, however, the recommended number of simultaneous swaps is four.
Step 1: Identify volumes to swap Symmetrix Optimizer identifies a pair of hypervolumes to swap based on recognizable patterns of hypervolume activity and criteria. Assume the red volumes have high activity and the blue volumes have low activity.
Figure 2. Identify volumes to swap
Step 2: Copy a volume to DRV Symmetrix Optimizer swap commands are passed to SymmWin, which will assign one DRV (DRV1) as a
third mirror for hypervolume A. A second DRV (DRV2) will be assigned as a third mirror for
hypervolume B. All tracks on the third mirror are marked invalid. Tracks are copied from the valid mirrors to the two DRVs. After the DRVs are synchronized, the two original swap physical mirrors are marked Not Ready (volume A and volume B) and their attributes are swapped (Figure 3). Both hypervolumes still have two (or more) physical mirrors. Host activity to the hypervolumes is now directed to DRVs and the other mirror.
EMC Symmetrix Optimizer A Detailed Review