Fragmentation over time
Since the X25-Mreviews, I run through our test suite in a specific sequence as to prevent fragmentation from negatively impacting each test.  Most drives combine writes these days, and a sequential write pass over a given area (or the entire drive) will clear the fragmentation caused by previous combined writes.

So what normally happens is a drive will do poorly after being beaten up by our testing, but snaps back to 100% after a single pass write puts everything back in order.  Samsung and X25-E drives do this.  With its firmware update, the X25-M sticks to its speed limit regardless of what you do to it.  The Vertex and UltraDrive are a different story entirely…

Indilinx based OCZ Vertex and Super Talent UltraDrive ME SSDs Reviewed - Storage 52

Hmm, that’s a bit under 170 MB/sec to say the least.  Write speeds had in fact dropped to an average of 42 MB/sec.  It started out much lower, but it appeared the drive was attempting to straighten itself out as it went.  Remember the first pass almost always looks bad (but not normally *this* bad).  The second pass usually comes out with things back to normal.

Indilinx based OCZ Vertex and Super Talent UltraDrive ME SSDs Reviewed - Storage 53

66 MB/sec average.  That seems odd.  The drive should have bounced back.  Lets try a few more passes…

Indilinx based OCZ Vertex and Super Talent UltraDrive ME SSDs Reviewed - Storage 54
Indilinx based OCZ Vertex and Super Talent UltraDrive ME SSDs Reviewed - Storage 55
Indilinx based OCZ Vertex and Super Talent UltraDrive ME SSDs Reviewed - Storage 56

The result was 74, 77, and 79 MB/sec.  So it seems even the newest firmware is unable to defragment itself on the next sequential pass.  I thought perhaps this issue was limited to HDTach, so I did a quick pass with ATTO bench:

Indilinx based OCZ Vertex and Super Talent UltraDrive ME SSDs Reviewed - Storage 57

Indilinx post-bench ATTO after attempting to recover performance manually.

This is bad news.  The only real edge the Vertex and UltraDrive ME had over the Intel drive was in write speeds.  It took us 5 full-drive sequential write passes to get write speeds up to 80 MB/sec.  The fragmentation also appears to be taking its toll on read speeds – it has trouble breaking 190 MB/sec (also seen in the HDTach passes above).  Hmm, 80 MB/sec sounds familiar:

Indilinx based OCZ Vertex and Super Talent UltraDrive ME SSDs Reviewed - Storage 58

X25-M post bench ATTO pass performed after a *single* pass write.  Note the different scale.

The X25-M, fragmented or not, reaches the same write speeds as the unrecoverably fragmented Indilinx drives.  It also does much better with small random writes, ramping up to full speed much earlier in the game, and doubling the small file performance of the Vertex and UltraDrive.

We worked with OCZ very early in the game on this fragmentation issue, before the Vertex went retail in fact.  Indilinx opted to keep write stutter as low as possible out of their 4-channel controller, and in doing so chose to not make any effort towards defragmenting flash blocks on the fly.  As a result, the best performance tends to settle out at about 80 MB/sec as seen above.  They did however have a bit of a trick up their sleeve…
« PreviousNext »