[OpenAFS] Overview? Linux filesystem choices

Gary Buhrmaster gary.buhrmaster@gmail.com
Thu, 30 Sep 2010 06:40:55 -0700

On Thu, Sep 30, 2010 at 05:19, Stephan Wiesand <stephan.wiesand@desy.de> wrote:

> Anyway, the next best option if ZFS is not available is to run parity checks on all your arrays regularly.

Perhaps it is the best one can do, but be aware that
a (rare, but real) failure mode of disks is that they
return the contents of a different block than asked.
No amount of background scrubbing fixes that
unless the failures are solid (and they are usually
not).  That does not even include the issue that
most disk controller data paths (and cache
memories) are not even parity checked, and
bit flipping can happen there too.

NetApp recognized this and dealt with it with
the WAFL file system years ago.  They actually
wrote a checksum for the block and the block id
onto disk and checked when they read a block

Getting back the data your wrote is a hard
problem.  ZFS presumes that everything
downstream of it will (eventually) fail.  There
is overhead there, but it does solve a set
of problems that other solutions do not.
(And the highly paranoid presume ZFS will
fail, so take different precautions).

As Jeff stated, if you really care about your
data, you need ZFS.