51% can be enough Was: Wiping a drive

[Alexander Klimov <alserkli () inbox ru>]

On Wed, 15 Oct 2008, Craig Wright wrote:
> Even at 92% per bit, the recovered data is useless and random. This
> is detailed in the paper mentioned before. At 49% - this is a modern
> drive - the toss of a coin is more accurate.

Not sure what are exactly these numbers, but if it is probability of
correct recovery than they are not necessary useless. Suppose you edit
a text document and your editor automatically makes a backup copy of
it every five minutes. Even if backup is done with the same filename,
with journaling filesystems you end up with many dozens of copies of
the file content on your disk.

Now, if locations of backups are predictable (the document is long
enough to make correlations sufficiently large), it is possible to
recover the document even if you can read every bit with 51% success
rate (btw, the probability cannot be less than 50%, because in that
case you should always guess the opposite) -- simply count what bit
value among copies is recovered more often.

Btw, the standard way to wipe disk on Linux is to use shred that is a
part of coreutils that are already installed on almost every Linux
system.

-- 
Regards,
ASK