Fedora Present and Future: a Fedora.next 2014 Update (Part I, "Why?")

[Chris Murphy]

On Mar 24, 2014, at 1:04 PM, Tom Horsley <horsley1953 at gmail.com> wrote:

> On Mon, 24 Mar 2014 11:30:23 -0700
> Rick Stevens wrote:
> 
>> It's not so much an issue that hard disks don't fail, but they usually
>> start giving some indication they're having issues
> 
> Not any disk I ever had fail. All of 'em worked perfectly
> right up to the instant where they wouldn't boot
> one day. Nothing in the logs, no smart warnings, nuthin.

I think it's something like 60% for the prediction of SMART failures and often you have maybe minutes to do something about it.  Anyway, SMART definitely permits a lot of drive failures without prior warning. 

That percentage goes up a lot if you look at all of the attributes, and choose to replace based on attributes trending with higher raw values (trending lower normalized value) that are associated with prefail. But I think it was the Google study that found even looking at all attributes still wasn't enough of the right kind of data to do really useful failure prediction.

> In fact, I've always wondered what good all the smart
> reporting is for since no disk I've owned has ever had any
> problems reported right up to the point where it wouldn't
> talk enough for smart to access it :-).

The SMART overall-health self-assessment test result is next to useless. It's basically a pass/fail. A better picture emerges when looking at the individual attributes. Still it's not fool proof because all you're doing is somewhat subjectively saying "OK if bad sectors start to trend upward at a rate greater than, say, 1 per week, I'm replacing the drive." I don't know that anyone has a clear line in the sand on the attributes though, there's a huge gray area.


Chris Murphy