[mythtv-users] ZFS - what's the scoop?

Richard Freeman r-mythtv at thefreemanclan.net
Thu Oct 18 19:04:35 UTC 2007


John Drescher wrote:
>>
> I did not have the ZFS raid layer activated I agree this may have
> something to do with the problem.
>

In theory ZFS with RAIDZ should perform better than ZFS on top of md.
The reason is that ZFS is designed to avoid modifying data within a raid
stripe - instead it always writes to unused parts of the disk (and
deallocates used parts of the disk).  This avoids having to read the
rest of a raid-5 strip for parity computation.

I've been following ZFS with great interest - a few comments:

1.  In theory it has a LOT of potential.
2.  It is still immature - maybe not quite ready for prime time
(especially the linux implementation of it).
3.  RAIDZ is a GREAT idea, but as far as I can tell it has no support
for reshaping the array - so it is currently more limited than md.

I know it is a layering violation, but the beauty of ZFS is that it can
do more BECAUSE it is integrated from top to bottom.  It can take
short-cuts that would be unsafe at any individual layer boundary because
it knows what is going on at all levels.  In my mind this is a good
reason for breaking layer boundaries.

Unfortunately, due to the license it will be a while before we see it on
linux...


More information about the mythtv-users mailing list