In article < vpadnVkUKdnZskvYnZ2dnUVZ_uejnZ2d@metroca
stcablevision.com>,
Bill Todd  <billtodd@metrocast.net> wrote:
>richard wrote:
>
>...
> 
>
>The disk manufacturers, at least, don't:  the environment for which
>their disks are designed is quite publicly specified, and it's hardly
>their fault if those who use the disks don't pay attention to it.
>
>Failure rates of (S)ATA units have been studied by users as well as
>manufacturers.

Both large scale users (computer and storage array makers) and disk
makers have very detailed data on this.  However, this data is not
shared with competitors, nor with consumers.

Several studies have been published in the open literature:

> while the real-world-environment failure rates
>encountered by the former (such as the Internet Archive project and the
>- I'm sad to say now apparently late - Jim Gray at Microsoft)

May Jim rest in peace.  I hope he died the way he wanted to.  At this
point, that's the best we can hope for.

In addition to his data, this year's FAST conference (go to
www.usenet.org and search for FAST2007) had two papers on real-world
disk failure rates.  The first one (by Bianca S. from Carnegie Mellon)
got best paper award, and has a lot of information about a variety of
settings.  The second one (from google) has some very interesting
information about a 5-year study; unfortunately, it mixes older and
newer disks.

The second paper has some astonishing data: namely that disks live
longer if you don't keep them too cold; 30 or 40 degrees are better
than 15 or 20.  We know that really cold temperatures (5 or 10
degrees) are bad for disks, but I was quite astonished by this result.
A very senior person from a disk manufacturer was sitting next to me
during this talk, and was shaking his head.  The above observation
quite bady violates many things we thought we had known, and might be
an artifact of mixing different disks that use different housings in
the same data set.  So don't take it too serious yet.

Please note that disk lifetime is also affected by other factors.
Such as workload: continuously seeking is bad; continuously writing
is also bad (as it increases the risk of off-track writes).  This is
particularly true for consumer-grade disks (which is often but no
always synonymous with IDE/SATA disks); those are typically specified
for 40-hour-per week operation, instead of 24x7.

Disk lifetime (and error rate) is obviously affected by temperature,
except that the recent Google result above confuses that issue.  It is
also seriously affected by vibration, in particular for consumer-grade
disks (again, often IDE/SATA), which can't simultaneously servo and
transfer data.  For this reason, it is important to use
vibration-absorbing disk mounts, low-vibration fans, and isolate disks
from other vibrating components (such as CD-ROMs and other disks).
Mechanical shock during operation can be very very bad, so don't kick
your computer just because your program doesn't compile.

>Perhaps you're just unlucky, or unusually hard on your disks, or the
>design of the cases you use sucks, or you're using older disks (the
>newer ones don't tend to get nearly as hot as you claim when properly
>cooled:  for the past several years my internal 7200 rpm Seagates have
>idled at around 20 degrees C. below their nominal 55 degree C. maximum,
>according to S.M.A.R.T., with no special cooling arrangements such as
>you describe:  the normal influx of air to the front of the drive bays
>caused by the PSU fan plus a single auxiliary 80mm. fan is more than
>sufficient for them).

Warning: 10K and 15K RPM SCSI/SAS/FC disks run considerably hotter
than 5400 and 7200 RPM IDE/SATA disks.  Particularly true for 2.5"
enterprise-grade disks.  All those disks should be equipped with a fan
that guarantees good airflow.

>In any event, don't presume to generalize from your own experience -
>especially given the ease with which you could find broader relevant
>information.

If you want to generalize from your experience, your experience better
be based on hundreds of thousands of disks.  Most home computer users
(fortunately) don't gather that kind of data.

--
The address in the header is invalid for obvious reasons. Please
reconstruct the address from the information below (look for _).
Ralph Becker-Szendy      _firstname_@lr_dot_los-gatos_dot_ca.us

[ Post a follow-up to this message ]