ATA failure with 4.6.2 & 250GB drive?

[Kevin Oberman]

> Date: Wed, 15 Oct 2003 13:08:05 +0100
> From: Scott Mitchell <scott+freebsd at fishballoon.org>
> Sender: Scott Mitchell <scott at fishballoon.org>
> 
> Hi Kevin,
> 
> As always, a most enlightening response :-)
> 
> On Tue, Oct 14, 2003 at 09:02:14AM -0700, Kevin Oberman wrote:
> > It's a real drive problem, but possibly not a terminal one. (I had the
> > same issue on one of my drives a few months ago and it's fine now.)
> ...
> > The fix/workaround is to move the file(s) involved so that the damaged
> > blocks are marked free and relocated to spar space on the drive. You
> > can try to figure out just which file(s) use those blocks. There
> > might even be a reasonable way to do this...I just don't know what it
> > is.
> > 
> > Another "fix"is to simply copy the drive onto another and then copy it
> > back. dd(1) will do the trick as will dump/restore. (I'd suggest the
> > dump/restore to copy the data out and dd to copy it back if the disks
> > have identical geometries.) Once the data is restored to the original
> > disk, the bad blocks will have been re-directed by the drive and will
> > no longer trouble you.
> 
> I'll probably pull it out and run the Maxtor diagnostics over it to see
> if they turn up anything interesting.  Do you think a low-level format
> would be useful?
> 
> Very annoying to have two of these things die almost immediately after
> installation.  Does not encourage me to buy more Maxtor products in the
> future :-(

FWIW, the drive I ran into this on was an IBM (now Hitachi). Back
blocks on a hard drive are unavoidable. Plating techniques are amazing
(especially when I worked for so many years with coated media), but
they are simply not perfect, especially with the magnetic domain size
on modern drives.

This is more likely an indicator of inadequate testing of the
drive. Normally the bad spots on a platter should be detected and
re-mapped before the drive is shipped. Every platter has bad blocks,
but they really should be caught before the unit is shipped. The issue
is marginal areas that pass the tests when the drive is first turned
on, but produce too weak a domain to be detected when the drive has
aged a bit. These should be caught before shipping, but will continue
to crop up during normal use as the heads age. As long as they are
detected during a write attempt, you never see them. But, if an
already written sector is becomes unreadable, there is not graceful way
to recover.

I don't know much about ATA disks, but they probably have ECC to
further reduce the chance of this happening, but failures still
happen.

I doubt that a low-level format will help any. The drive diagnostic
should re-map the bad region and the drive may continue to operate
well for years. The sign of impending doom is when this thing happens
repeatedly. That usually means that the disk's days a near the end.

This is from someone who was very involved in disks back in the 70's
and 80's, but who has never worked much with IDE/ATA disks and whose
major experience was with SMD drives. I like to think I still know a
little about them, though I may simply be growing senile.
-- 
R. Kevin Oberman, Network Engineer
Energy Sciences Network (ESnet)
Ernest O. Lawrence Berkeley National Laboratory (Berkeley Lab)
E-mail: oberman at es.net			Phone: +1 510 486-8634