[ale] Recommendation for off-line storage?

[James P. Kinney III]

> At 03:48 PM 5/10/2006, Greg Freemyer wrote:
>
>>If so, there is NO data or gaurentee on disk drives maintaining data
>>with power off.

I have to disagree with this. The greatest data loss issue with any
magnetic medium is the presence of an unplanned magnetic field that
changes the bits on the medium. It doesn't matter if it's tape or hard
drive, the effect is the same.

The hard drive platter is made from a very stable material that hold it's
magnetic domains intact for very long times. While all magnetic domain do
tend to spread over time, there is nothing about powered on status that
prevents this occurance.

Hard drives fail in use due to mechanical issues (platter bearings,
armature mechanicals, motor, etc), electrical (control electronics board
fails) or platter meduim failure. The mechanical failure rate can be
attenuated by infrequent use which reduces the strain on components.
Electrical component failure seems to be of two types: sudden failure
related to spike damage and manufacturing errors that cause poor joints,
poor components, and improper signalling.

The latter failure, platter medium failure, is more complicated. It
involves the manufacturing aspects of the physical dimensions of the disk
(basically it's flatness, it's roundness and it's uniform density) and the
uniformity of the magnetic medium coating. The physical characteristics
will have a great effect on the mechanical longevity of the drive. Out of
balance spinning objects put excessive stress on mechanical systems.
Flatness errors can lead to head impacts with the disk (which are nearly
always catastophic). The magnetic medium is quite stable in use and in
storage. Manufacturing errors can have coatings that are "thin" and this
often results in bad sectors. Writing to the platter has more of an effect
on the magnetic medium as the heads magnetic field exerts a mechanical
force on the particles in the medium. Reading is passive and exerts no
force on the medium. Since hard drive are not handled like RAM (they don't
get refreshed periodically) the long term storage has no detrimental
effects on the medium itself. The magnetic dispersal rate is comparable to
that of high-speed tapes.

What does damage hard drive is mechanical shock of moving between storage
and active use. During mechanical shock, a phonon wave is propogated
throughout the entire hard drive. Just like waves in a water tank, phonon
waves can be reflected and refracted and they can be constructively and
destructively combined. When the wave moves through a magnetic material,
the magnetic fields are reoriented at the atomic level for a brief time.
When the wave has passed, the material "relaxes" back to it's undisturbed
state. Almost. As the wave passes through, the alignment of the magnetic
domains is changed. This causes internal stresses that affect the final
relaxed state. So the magnetic m edium gets altered with each passing
phonon wave. The stronger the phonon wave, the greater the overall
disturbance. Given enough of a shock, the entire magnetic domain for a
cluster can be distrupted enough to render it unreadable. Magnitize a
sewing needle by rubbing it on a magnet. Then strike it with a hammer and
notice it looses it magnetic field.


>>
>>We do it a lot around here, but we also make a tape backup before we
>>put the drives on the shelf.
>
> Well that certainly changes the equation.  I was indeed planning to leave
> these drives on the shelf.
>
> I was planning to make two copies before deleting the on-line data -- so
> maybe I need to send one copy to tape (for long-term retention), and one
> to disk (for less-reliable retention, but easy searching and restoration).
>
> So now I'm looking for a tape-based solution too?  Same parameters apply,
> I suppose:  Linux-friendly, non-proprietary, rack-mountable, minimal
> hassle.
>
> Although the last time I dealt with tape drives, the tapes were only 20-40
> GB.  I have 400 GB and the data center is a remote facility.  Now I'm
> talking a tape carousel?  Am I going about this all wrong?  Should just
> make two copies on IDE drives, take them off-line, and hope for the best?
>
That would work just fine.

> A----
>
>
>
> _______________________________________________
> Ale mailing list
> Ale at ale.org
> http://www.ale.org/mailman/listinfo/ale
>