Performance Over Time and TRIM
Performance Over Time:
While we saw no significant degredation during our test sequence, we did note some some oddities during the sequential write passes we perform after testing has completed. We do these tests on every SSD we test, but only present them when we see things out of the ordinary:
Normally the first sequential write pass acts to defragment the flash area of the SSD, causing the second write pass to smooth out and return to like-new. With the M6M we saw a bit of weirdness. First, the write speed (dark red) did not completely clear itself up after the first pass. Second, the random writes conducted during our sequence appeared to slow the reads across the entire span (bright red, first graph), but this did return to full speed after the first pass write completed (bright red, second graph). Additional passes did not seem to clean up the write speeds any further, so I let the M6M idle overnight in case we were interfering with some background garbage collection routines the drive might have been trying to execute. After an overnight idle period, we saw this result:
Well that was unexpected. The write performance has decreased even further.
TRIM:
Suspecting that the M6M may be heavily optimized for TRIM, we hit it with a quick format. Well, we thought it would be quick:
The 'quick format', somthing that is near instant on an SSD, took several minutes to complete on the M6M this time around. This was abnromal even for the M6M – the initial formats of this sample were very quick. The difference here was that we had written a bunch of data in an environment where TRIM was not actively being used, and the subsequent format operation forced the M6M to 'catch up' and TRIM its entire flash area. The end result was the expected as-new condition:
Conclusion: It appears Plextor tuned the M6M heavily around the use of TRIM. Running it in an environment without TRIM (IDE mode / some RAID configurations) may very well lead to a noticable hit on performance as the drive is written to over time. It may also negatively impact the performance of VM images stored on the M6M, as VM images are static files that are randomly modified in-place, and therefore not TRIMmed by the host operating system.
As always Allan, great
As always Allan, great review. In the day and age of multiple storage options, I do have a hard time differentiating results in the tests provided. For relevance sake, and this is only a question, try and reduce compared drives for easier understandability. As in top 2 direct corporation drives (by connection type and $/Gb) and one of the others; ie SATA, mSATA, PCIe, M.2. With the ever expanding storage market,It’s getting flooded,and just want easier to discern choices. In any case, you house at PCper rock! Keep up the excellent work and keep asking question as to what happens next.