Not all SSDs are Created Equal

Results of the SNIA PTS on Seventeen SSDs and one HDDSSDs vary widely in performance.  This is something that becomes amazingly clear when a number of these devices are put through a battery of tests.

Calypso Systems ran the SNIA SSD Performance Test Specification (PTS), outlined in an earlier post in this blog, on seventeen SSDs and a single HDD.  The results appear, in miniature, in the graphic for this post.

The chart shows the random 4k byte write IOPS (input/outputs per second) on the vertical axis for these eighteen drives (across the horizontal axis) for varying block sizes (the depth axis.)   Smaller blocks are toward the rear of the chart, with larger blocks toward the front.  The blue bars are SLC SSDs (single-level cell flash), the pink bars are MLC SSDs (multilevel cell flash), and the yellow bars are the HDD.

A few quick observations can be made from this chart, no matter how tiny it is:

  1. SSDs tend to perform better with small block sizes.  In fact, most are optimized for 4k – 8k byte blocks, and show a clear peak at this range.
  2. In general SLC SSDs have higher IOPS than their MLC counterparts, and both SLC and MLC generally perform better than the HDD.
  3. But not all SLC SSDs outperform MLC SSDs.  It is easy to see that some pink series are significantly higher than the lower blue series.
  4. It is harder to tell, but the HDD performance varies far less than does the performance of the SSDs, although its very low IOPS make this more difficult to discern.
  5. The lowest-performing SSDs have lower performance than the HDD.

This last is a bit of a surprise, since it is generally understood that any SSD will run faster than any HDD.  While this is the case for a new SSD, after some period of use it may no longer hold true.

In the post about the SNIA PTS we explained that SSDs suffer performance degradation with use, and the figures in this chart, measured using the PTS, are run on SSDs that have been preconditioned until they reach a steady state.  The steady-state IOPS of the lowest-performance SSDs in the chart falls below the steady state IOPS of the HDD.

Significantly more detail on these tests is available in the report: Putting SSDs to the Test, jointly written by Objective Analysis, Coughlin Associates, and Calypso Systems.  The report can be purchased directly on the Objective Analysis website.