DS2246 disk shelf increases performance density by 60%. How is this calculated?

[wian]

The DS2246 disk shelf increases performance density by 60%. Lets open up the discussion to hear from our community experts on how this is calculated.

[vmsjaak13]

Let's say:

125 iops for a 2.5 inch 10k SAS drive.

175 iops for a 3.5 inch 15k SAS drive.

DS2246 2U = 24 * 125 iops = 3000 iops = 1500 iops per U
DS4243 4U = 24 * 175 iops = 4200 iops = 1050 iops per U
DS14mk4 3U = 14 * 175 iops = 2450 iops = 817 iops per U

Regards,

Niek

[woods]

Here's how the estimate for 60% greater performance density (IOPS per rack unit) was calculated.

A NetApp storage system configured with the DS4243 disk shelf with 16 x 15k RPM, 3.5" SAS disk drives as tested under an OLTP workload.  The number of drives was chosen to ensure that the storage subsystem, not the controller, would be the performance bottleneck.  Data was collected plotting latency (ms) as a function of the load (IOPS).  Then the same test was run with everything the same except the backend was changed to a DS2246 shelf with 16 x 10k RPM, 2.5" (SFF) SAS disk drives.

A plot of the results showed that the 10k RPM drives provided 78% to 82% of the IOPs of the 15k RPM drives across a wide range of fixed latencies. Hence, the 10k RPM drives provided ~80% of the performance of 15k RPM drives.

Performance density was determined by factoring in the rack space needed for each set of drives. The DS2246 disk shelf packs twice as many 10k RPM drives into the same space as the DS4243 shelf with 15k RPM drives. So, 80% of the performance (IOPS) in 50% as much space (rack units) translates into 60% higher performance density.