Backblaze has published the first SSD edition of its regular drive statistics report, which appears to show that flash drives are as reliable as spinning disks, although with surprising failure rates for some models.
The cloud storage and backup provider publishes quarterly and annual Drive Stat reports, which focused exclusively on rotating hard drives until last year. Backblaze said it will initially publish the SSD edition twice a year, but that this may change depending on how valuable readers find it. The 2021 Drive Stats report was published in February.
In a blog post detailing the latest probing, Backblaze cloud storage evangelist Andy Klein said the SSDs are all used as boot drives in the firm’s storage servers, and that Backblaze only began using SSDs this way from Q4 of 2018. He pointed out the drives do more than just boot the servers, they also store log files and temporary files produced by the servers, and so each SSD will read, write, and delete files depending on the activity of the server during the day.
At the end of 2021 Backblaze had a total of 2,200 SSDs in its storage servers, but these had been built up from zero starting in Q4 2018, and so some have been in use for longer than others, and some models are also represented by a small number of examples which can skew results.
As an example, Klein singled out the Crucial CT250MX500SSD1 and the Seagate ZA2000CM10002, which were found to have high annual failure rates of 43.22 per cent and 28.81 per cent respectively.
However, there were only 20 Crucial drives, all of which were installed in December 2021, and just four Seagate drives, one of which failed in early 2021. In both cases, the annual failure rate calculation is based on very little data. Backblaze generally expects to see a 1-2 per cent annual failure rate, and “anything less is great and anything more bears watching,” Klein said.
Looking at the annual SSD failure rates for just 2021, Klein pointed out there is an extremely wide confidence interval for each drive. He stated that Backblaze considers a respectable confidence interval to be less than 1 per cent, with 0.6 per cent or lower being optimal. Only one SSD model meets the 1 per cent cut-off, which is the Seagate ZA250CM10002, although another Seagate model (ZA250CM10003) is close.
Looking at the cumulative annualised failure rates for all SSDs since the start of 2019, Backblaze finds that the curve on the resulting graph fits nicely below the target 2 per cent mark, whereas the quarterly failure rates showed spikes at or near 3 per cent for some quarters.
However, the firm warned that its total SSDs in service is a relatively small number of drives on which to start basing any conclusions.
“That said, the more data the better, and as the SSDs age we’ll want to be even more on alert to see how long they last. We have plenty of data on that topic for HDDs, but we are still learning about SDDs ,” Klein said.
Looking at three of the older SSD models that Backblaze has in operation, the firm found that all three had cumulative annualized failure rates below 1 per cent for 2021, which compares with the cumulative AFR for all SSD drives of 1.07 per cent.
However, while one SSD type (Seagate ZA250CM10002) appears to be following a classic “bathtub” failure curve – experiencing early failures before settling down to an AFR below 1 per cent – the other two drives apparently showed no signs of early failure and have only recently started to fail, a mode Backblaze said is similar to that it has seen with hard drives it has in operation that no longer fit the bathtub curve model.
For comparison, the cumulative annualized failure rate for Backblaze’s large number of installed hard drives was 1.4 per cent, as noted in its 2021 Drive Stats report. Klein was keen to stress that the two groups of drives cannot be considered to be at the same point in their life cycles, and the firm will continue to build its comparison of SSDs and HDDs over time as it gathers more data on the two groups.
The drive data is made available by Backblaze for anyone to download and analysis for themselves. The data contains figures for two extra drive models, Samsung 850 EVO 1TB and HP SSD S700 250GB, which are not included in the report because they were used by Backblaze for testing purposes.