[vox-tech] Which distro for file/print/web server

[Rick Moen]

Quoting Bill Broadley (bill at broadley.org):

> While technically true, often eSATA is combined with a multidisk chassis
> and has a lame/broken chip that multiplexes a single SATA connection to
> multiple drives.  Said lame/broken chip often hides the SMART data.

Yes, definitely don't buy those.

> I find is similarly frustrating when the RAID controller does the same
> thing.  It's really really annoying to have to pull a failed drive to
> get it's model and serial number so you can RMA it.

Well, sometimes there are compensating advantages to the RAID
controller, and ways to query the latter to determine the model and S/N
of the failed drive.  Not that I'd recommend one, but my old firm VA
Linux Systems sold a lot of Mylex RAID HBAs that met that description.

> The studies that I've seen show SMART is useless for predicting
> failures.

That depends on what you mean by 'predicting failure'.  You can use the
SMART data to determine how many hours of online use the drive has had,
and replace it before it approaches MTBF.  You can also note the pattern
of accumulating errors and reallocations, a sharp rise in which
correlates well with some failure modes albeit not many others.  (The
Google paper notes this fact.)

Frankly, my main concern is many HD manufacturers' tendency to make
their drives lie when reporting such data.  I have my hunches about
which manufacturers and model lines are most guilty of this lying, but
can't prove it.

Alternatively, you can just RAID1 drive pairs, and deal with failures as
they arise without particular worry in advance.