0
What is the average life of a SATA hardrive?
Almost all data I can find gives failure rates for the first 0-5 years, but none seem to actually find the end of the life of the drives.
The reports, charts, and studies by google, backblaze, and the likes only tell part of the story as they focus on the first 5 years +/-.
Hypothetically to say 50% of drives die in 8 years does not infer the other 50% die in 16 years. Is there a chart that takes 100% of a set of drives to their death and gives the results? Or something that would provide equivalent information?
Assuming heavy consumer work load on consumer drives in typical climate controled home/office, what is a real world average of hard drive lives? Again, not failure rates given a (short) set life span.
Real world results for us is we've had less than 10% drive failure in 10 years and never failures close together so I am pretty comfortable with using aged drives but like to be informed where possible; Our current set of drives range from 0-8 (running) years averaging probably around 3-4 years, the most recent failure was a 5 year running drive. Further We have a 40gb and 80 gb drive the each are well over 10 years (manufacture date) old and still get used reliably here and there. Enough data to say SATA HDDs last reliably well beyond 5 years, but not enough to show a trend of how long.
Backround:
We are moving to an OBR10 setup for a small business with aged SATA drives of 4-6 years and I am trying to figure out how prudent it would be to move to a 3 copy MD RAID 10 vs 2 copy.
With daily data mirrors and full backups it would not be detrimental to have a full primary array loss and need to rebuild and restore from backup, but I would love to avoid such a scenario. However I cannot seem to find data that looks well beyond the age of our current drives. and their is no indications that they fail in droves at the 5 +/- year mark where the data seems to stop.
3
They tend to fail by usage, not age. So it really really depends on workload. The best data I can think of is that which is provided by backblaze. https://www.backblaze.com/blog/hard-drive-failure-rates-q2-2016/ - No one else I know of publishes anything near this.
– djsmiley2k TMW – 2017-01-02T07:57:06.4232@djsmiley2k Annualized failure rates for drives 0-5 years old have nothing to do with the average life span of drives and further nothing to do with rates of failure after 5 years. I would agree that life span without corresponding annualized failure rates for a given group is also problematic for making decisions, but where is the data for 5-10 years+? There has got to be many millions of drives, if not billions older than 5 years still running reliably. My assumption is someone somewhere has some insight. – Damon – 2017-01-02T08:09:08.400
Also the other thing to bear in mind is if you've got a 10 yr old drive, that's likely running SATA 1? At some point it becomes harder to pick up replacement drives 'on the spot' so to speak and also more expensive (if required) to recover data from said drives. – djsmiley2k TMW – 2017-01-02T08:09:59.680
@djsmiley2k I'm not sure I agree. SATA I still works on new hardware (SATA III) and they have slowed down on switching things up so often. Not to mention replacing an old failed drive with a new drive on the new SATA interface and adding it to the array is not a problem; further we do not need to find a drive of the same vintage to mitigate a failed drive so no problem there. Also, data recovery would not be needed with mirrors and backups. – Damon – 2017-01-02T08:22:35.820
then the question arises 'why do you care if the disk is going to fail?' – djsmiley2k TMW – 2017-01-02T08:31:16.147
@djsmiley2k namely downtime, possibility of user error during restore, mitigating time array spends in degraded state. We do care about the data, we just work with a budget, like to make informed decisions, and currently have no information on the question at hand. – Damon – 2017-01-02T08:38:08.750
Let us continue this discussion in chat.
– djsmiley2k TMW – 2017-01-02T08:50:56.440Ok, Car Anology time - there's cars from the 1920's that still run. Would you trust yourself to them? Also the fact there's no 10 year data is due to the reasons I pointed out, the people who test on large enough scale to generate this data (google, backblaze) don't run the hardware that long, because it doesn't make sense to do so. Interfaces and technolgy changes mean they move to newer versions before the hardware ever reaches that 10 year point. – djsmiley2k TMW – 2017-01-02T10:49:54.823