up vote 8 down vote favorite
1
share [g+] share [fb]

is it possible to test a hard disk and predict when it will die on you (roughly)?

link|improve this question
76% accept rate
5  
lol! :D you can test it by formating/overvriting/copying bilions same files and when it dies you will know! :P – FeRtoll Feb 8 '11 at 21:01
8  
Your question would make a good Masters/Ph.D project. – aqua Feb 8 '11 at 21:42
No answer here will be correct if you use a hammer. – M28 Mar 7 '11 at 3:21
feedback

migrated from stackoverflow.com Feb 8 '11 at 21:07

This question came from our site for professional and enthusiast programmers.

5 Answers

up vote 10 down vote accepted

While @Alex Howansky's answer made me chuckle endlessly, I think what you're really looking for is something like SMART diagnostic data, which is a predictive failure tool. Any modern drive that supports it will WARN you if it thinks something is wrong based on a pattern of small failures...major head crashes are still basically impossible to predict, though.

link|improve this answer
1  
SMART is not predictive, it just tells you when something bad has already occurred. Google's survey of hard drive failures in their server farms found that SMART data did not corrolate with actual failures. 56% of drives that crashed didn't show SMART failures. – Blackbeagle Feb 8 '11 at 21:35
3  
@Blackbeagle - SMART is predictive in that it shows indicators of general drive health. If you see raw read error count increasing, that's a sign that you're experiencing problems with the drive and you should replace it before it fails - predicting a failure. Unfortunately a lot of failures on mechanical drives, as I already said in the answer, cannot be accurately predicted like this; I haven't seen those numbers you claim but they sound exactly right to me. But saying that SMART isn't designed with the intent of failure prediction is flat out wrong. – Shinrai Feb 8 '11 at 21:45
While SMART may have been designed with failure prediction in mind, Blackbeagle does have a good point in citing Google's finding that in reality, SMART data did not reliably predict failure. I presume that the OP is more interested in reality than theory. :-) – skypecakes Feb 8 '11 at 22:29
1  
@skypecakes Agreed, but again...you can't predict catastrophic failures with any reliability, period, so in the case the answer is "You're SOL." Something is better than nothing in my book. – Shinrai Feb 8 '11 at 22:30
feedback
up vote 7 down vote

Sure, your hard drive will die shortly after you hear "bzzzzzzzzzzz grrrr grrrr grrrr click click click DING" come from it.

link|improve this answer
11  
And at that point you can just subtract a small amount from your timestamp and you have the answer. – Gareth Feb 8 '11 at 21:04
12  
I can't really +1 this in good conscience, but I just want you to know I love this answer. – Shinrai Feb 8 '11 at 21:12
feedback
up vote 7 down vote

The S.M.A.R.T. tools can help display drive health information, e.g.

sudo apt-get install smartmontools
smartctl -H -A /dev/sda
etc.
link|improve this answer
2  
Assuming you're using a Debian-based Linux distribution, of course. – oKtosiTe Feb 8 '11 at 21:19
1  
Yes, apt-get assumes Debian/Ubuntu, otherwise it's yum install... or urpmi ... or pacman ... or emerge ... or yast2 ... etc. – Mikel Feb 8 '11 at 21:29
1  
I think @oKtosiTe's point was "not everybody runs Linux". ;) – Shinrai Feb 8 '11 at 22:41
Well, I could provide installation instructions for every possible platform, but that would be going over the top. The OP didn't say what platform they were using, so I think they can just follow the Wikipedia link or Google it. ;-) – Mikel Feb 8 '11 at 23:03
feedback
up vote 1 down vote

Not really. While it is possible to monitor some aspects and make predictions based on statistics, there are risk factors that can lead to instant failure without any warning. As the worst case scenario can strike at any time, I'd just plan around it, and not differentiate between failure modes.

So to guard against hardware failures, set up a RAID6 and swap harddisks if the controller tells you they are no longer usable; this protects reasonably well against typical failure modes (total unannounced loss of an entire disk and individual unreadable sectors on single disks), and for everything else (lightning strike, ...) there is your off-site backup.

I've since moved my entire storage at home to a single machine with 4 x 2 TB disks, which gives a net capacity of 4 TB, allows up to two disks to fail (this is an important feature, as statistically it is very likely for another disk to fail during a rebuild, and even single unreadable sectors count as failures if you have no redundancy left), and can easily be extended by adding more disks; move that to the broom closet and use diskless systems everywhere else, and receive a massive Spousal Acceptance Factor boost for future hardware purchases. :)

link|improve this answer
You'll need to register your SF account in order to regain that question on iSCSI – random♦ Feb 24 '11 at 19:20
feedback
up vote -1 down vote

Please visit following site: http://hddlife.com

There you will find factual information.

Would the supplier of the statement regarding google's experiences please quote an original source?

link|improve this answer
I looked into it myself, looks like labs.google.com/papers/disk_failures.pdf – Shinrai Feb 9 '11 at 14:47
feedback

Your Answer

 
or
required, but never shown
discard

Not the answer you're looking for? Browse other questions tagged or ask your own question.