Mailinglist Archive: opensuse (3996 mails)

[Danny Sauer <suse-linux-e.suselists@xxxxxxxxxxxxxxxxxxxx>]

Greg wrote regarding 're[2]: [SLE] IBM hard drive' on Fri, Sep 03 at 17:02:
> 
>  >>  I used to work at a large community college where I supported several
>  >>  hundred Gateway 2000 machines, both in labs and as instructor
>  >>  workstations.
>  >>  They all used WD drives.  I replaced a hard drive about once every 2-3
>  >>  weeks (not to mention the other crap hardware Gateway uses).
> 
>  >>  I will never own a Gateway 2K computer or another WD drive.  Others may
>  >>  have different experience, but my unique background formed *my* unique
>  >>  opinion. :)
> 
>  >>  --Danny
> 
> I'm not hardware reliability expert, but with "several hundred" machines only one every 2-3 weeks sounds pretty reliable to me.
> 
> ie.
> MTBF = 5 years for a good disk drive
> = 5 * 52 wks
> = 260 wks   (MTBF per drive)
> 
> You had "several hundred" machines, so the projected failure rate should have been closer to one per week.

MTBF = *Mean* Time Between Failures.  The Mean is also known as the Average.
While it's reasonable to assume that half of the drives will be dead before
that 5 years and half after, it's not reasonable to assume that they'll die
at constant intervals, one per interval right up to the point where they're
all dead.  I'll grant that, as the time in operation approaches the MTBF,
the odds go up that a drive will fail.  Since each drive's failure is
independant of each other drive's failure, though, you can not make any
accurate predictions on the group as a whole.  You can only predict the odds
that "a" drive will fail within a given time.  The odds that "a" drive will
fail within a week or so are pretty slim if quality control is good, because
the drives will be pretty consistent.  Consistency tends to pull the
distribution in towards the mean.  If we had a graph of this 5 year MTBF
drive, most of the drives *should* fail somewhere close to the 5 year mark.

Given that we had lots of drives failing at a pretty constant rate, and that
the constant rate was well before a reasonable MTBF, then one of two things
can be inferred.  Either 1) WD Quality control is bad, 2) we got most of
the drives on the bottom side of the MTBF graph (note that just as many would
have to exceed the MTBF in order to pull the mean up), or 3) the MTBF for WD
drives is lower than I think it should be.  None of those cases are good
excuses, IMHO, and therefore I don't use WD anymore - faulty probability
calculations or not.

--Danny