Backblaze sees drive failure rates tick up, asks if AI can help

Backblaze has issued the latest report detailing failure rates for the multitude of drives that power its storage and backup services, and is looking at recent trends in the figures as well as considering whether AI might lower those failure rates. As a storage service provider, Backblaze monitors an entire fleet of drives of …

[Flocke Kroes]

One obvious cause of failure

Before my beard turned gray hard disks arrived in corrugated cardboard boxes with over 2cm of foam protecting each side of the drive and the drive was in a sealed anti-static bag. The last four I received arrived in thin card envelopes with no protection at all. The good news is that one was so busted it failed smart so I could a clear reason to send it back. The other three 'work' but with over 10x the read error rate of much older drives. I have been hoping they would fail smart before the warranty expires but so far no such luck. So far I have only spotted one data corruption event but that is enough to destroy my confidence in all three.

The next distributor that does this to me gets the drive back unopened and I send a photo to the manufacturer.

[Missing Semicolon]

Re: One obvious cause of failure

I received a WD drive a while back like this. I sent it straight back, citing the WD guidelines for drive packaging for their warranty return process. It was refunded without quibble.

[Ken Moorhouse]

I would have thought that...

I would have thought that it would be important for the failure mode of each drive to be recorded (e.g., failure of the motor, failure of the platter, failure of the arm), then there's how many sectors that have been read/written throughout its life. If for example any activity gets near to the mechanical oscillation frequency of a component which causes amplification of that activity then surely that will reduce component life? Even the siting of the hard drive within a cabinet could affect life (heat, vibration) - I once had a washing machine which was sited in a non-ideal position. Every so often a part which had a turbine-like appearance within the machine would shear completely. I suspected this was due to the type of vibration the unit was subjected to. Various engineers told me that this was a very unusual component to fail, particularly in the way that it did.

Then there is the way hard drives are formatted. Surely the distance needing to be travelled to pick up the next sector of data would affect reliability, just as much as it affects performance? And cacheing must have an effect on reducing wear-and-tear.

Backblaze seem to be making life hard for themselves by treating each unit as a black box. Even the SMART data for each drive, over time should give some trend data to help with their mission. But then what do I know about hard drives?

[Shrek]

Re: I would have thought that...

For what it's worth, the raw data does seem to contain the SMART data. I think at the level of the blog post treating them as a black box is good enough to engage the reader.

[An_Old_Dog]

Not-S.M.A.R.T. + (Fake) AI

S.M.A.R.T. was supposed to be our pre-drive failure saviour which would tell us all about our drives' health and condition. But, many of the parameters are unspecified, or defined in some secret, manufacturer-specific way.

Of all my drives I've looked at via smartctl, the only values I saw which correlated to incipient (experienced) disc failure were the reallocated sector count and the recalibration attempts count.

That's useful, but not the superwonderful state of affairs we were promised.

I don't see how (fake) AI is going to improve things, given it will be looking at many ill-defined, and/or contradictorially-defined parameters.