How does Symmetrix Optimizer work?
The analysis methodology and the swap procedure are the core technology and the two main components of the Symmetrix Optimizer solution. The following sections provide a brief description of them.
When Symmetrix Optimizer evaluates performance statistics, it evaluates potential volume swaps based on how well they would improve overall performance. This analysis is based on minimizing disk service time (rotational latency plus seek time plus transfer time). Symmetrix Optimizer uses three strategies when determining which swaps to make:
First, swapping highly active volumes to disks with lower activity evens out the load across the physical drives in the Symmetrix system. This decreases the contention on the individual physical disks, improving the performance of both highly active volumes and low-activity volumes.
Second, swaps that relocate highly active volumes so that they are closer together on the physical disk will decrease the seek distance for I/Os to these volumes. When Optimizer swaps volumes, it will try to achieve this type of configuration since it decreases the overall I/O service time.
Finally, Optimizer tries to swap highly active volumes on the outer zones of the disk. This is because volumes located on the outer zones of the disk have faster transfer speeds.
Symmetrix Optimizer’s algorithm uses detailed disk performance information that takes into account several drive characteristics such as those gig-to-gig seek times, Zone Bit Recording, and bandwidth data.
Figure 1. Optimization strategies
By default, the goal of the Symmetrix Optimizer is to minimize the average service time of the Symmetrix disks. Optimizer Revision 4 introduced a new mode: Hot spot analysis. In this mode, the target function was changed to lower the peak service times rather than the average service times.
Optimizer only looks at back-end activity; it uses the back-end logical volume statistics obtained through a unique Optimizer system call. The metrics used are:
DA logical volumes reads
DA logical volumes writes
Logical volume prefetch
DA logical volumes blocks read
DA logical volumes blocks written
While modeling the disk service time, different weights are assigned to read, write, and prefetch activity. The assumption is that since writes are done as a background process, they are done in sequence and hence are more efficient. Optimizer does not try to follow DMSP policies; instead, it assumes that reads are equally spread among all the device mirrors, and all mirrors do all writes.
EMC Symmetrix Optimizer A Detailed Review