Backblaze shares 2017 HDD reliability stats

[Iota]

RAID is about resilience and maintaining uptime allowing a failing drive to be swapped out with minimal service interruption (the exception is RAID0 which just doubles the risk of data loss as failure of one drive can affect the data on both). They are not backup substitutes.

SSD don’t alter that - in fact they increase the need for backup because when they fail, they are (at present) more likely to fail catastrophically rather than giving early warning signs.

Oh don't get me wrong, I'm well aware of the reasoning for RAID for resilience and maintaining uptime, especially in a database use scenario amongst other uses. Equally it definitely isn't a substitute for regular backups as you pointed out. As for RAID 0 increasing the risk of data loss, that's something I'd challenge (regardless of the mathematics), a single drive dying is just as likely as a single drive working alongside another dying.

Someone linked this recently - https://techreport.com/review/26523/...-to-a-petabyte, although SSD may fail catastrophically, of those tested, they all surpassed their endurance specifications with ease, SMART reports the status of the drive (which can be easily be read in programs like HWInfo.). As long as you're not expecting to keep going past the endurance specs, I wouldn't imagine they're any less reliable than spinning plates of rust (in my albeit anecdotal experience so far, they're more reliable).

[peterb]

WRT to RAID0 my reasoning is that if each drive has an my mtbf of (say) 100 hours then the mtbf of 2 is 50 hours and as failure of either will probably result in the loss of data, the risk is doubled.

However I will add that stats is not my strong subject so I’m open to having my logic challenged!

[DanceswithUnix]

WRT to RAID0 my reasoning is that if each drive has an my mtbf of (say) 100 hours then the mtbf of 2 is 50 hours and as failure of either will probably result in the loss of data, the risk is doubled.

However I will add that stats is not my strong subject so I’m open to having my logic challenged!

I don't remember the maths being that easy, but two drives failing won't the the same as the chance of a single drive failing. Consider a RAID 0 of 10 drives, you just wouldn't, yet no number of drives gives you a certainty of failure. Not even with Seagates

[Iota]

WRT to RAID0 my reasoning is that if each drive has an my mtbf of (say) 100 hours then the mtbf of 2 is 50 hours and as failure of either will probably result in the loss of data, the risk is doubled.

However I will add that stats is not my strong subject so I’m open to having my logic challenged!

I've seen that argument (and the mathematics behind it), however logic dictates that if you have a single drive with an MTBF of 1.2 million hours, the MTBF of two single drives is still 1.2 million hours per drive and the MTBF isn't actually reduced when using together. However the chance of a drive failure has essentially increased by increasing the number of drives regardless of the MTBF.

Personally, I'll just go with the MTBF that has been tested by the drive manufacturers, they'll have tested the average length of time before a drive fails and provided a warranty based on that the average. Bearing in mind the variables involved with either an HDD or SSD failing (such a silicon variances for example), I doubt we will ever have absolutes provided.

I mean I could go buy lots of lottery tickets, while the chances of me winning have increased, it doesn't mean I'm going to. It also doesn't mean I'm not going to.

[Millennium]

I'm splitting my RAID tonight. I don't need the extra speed and just ran it because of E-PEEN from the old Athlon 64 days when SSDs were not around. If I need speedy storage I'll just get another SSD in future. Not worth the risk.

Backing up

[mikerr]

SSD don’t alter that - in fact they increase the need for backup because when they fail, they are (at present) more likely to fail catastrophically rather than giving early warning signs.

Will data recovery companies do anything at all for failed SSDs ?

I thought flash memory failures would make the disk go read only (I've had SD cards do that) - but every SSD failure I've seen has been a total death / non-detection (controller failure?)

[gagaga]

Will data recovery companies do anything at all for failed SSDs ?

Think some are, but it's hard to do data recovery on SSDs. The data is scattered across multiple flash chips - the 10% redundancy means random blocks scattered across the disks will be in the redundancy, and to add to the fun, 256bit on-disk encryption is pretty much standard with the encryption key buried somewhere in silicon in a place where it is meant to be unreceoverable.

So yeah, backup is important. Good thing is stuff like Acronis runs fast on SSDs, so twice a day backups aren't an issue.