Re: ZFS the perfect FS? if only...

From: Peter Schuller <peter.schuller_at_infidyne.com>
Date: Mon, 14 May 2007 20:50:07 +0200
> I believe raidz2 means two parity blocks so if you want 50%
> parity you'd need raidz4.  But that nit aside, you have a
> valid point.

Yes, sorry. I was kind of continuing a train of thought I have had
previously that was not specific to raidz2.

> It is not obvious at all that performance would not be a goal
> for a freebsd user!

Of course most people care somewhat for performance, but sometimes you
truly are not really that interested in getting better performance than
say a single disk. That is, performance is not always the goal of
setting up a raids, even if it will always be a nice bonus.

> It *is* obvious that you would want more space but not obvious
> how to do the conversion from an N disk raidz2 array to N+1
> disk raidz2 array *without bringing the whole array down*.

I realize it is not trivial to implement; but I do mean that even if
performance is adversely affected for an extended period of time, being
able to *do* the conversion at all is often very useful.

And even if it has to be be brought offline, that's still better than
not being able to do it at all...

> Even you may care about the array being down for hours/days!

... but yes, that will be annoying in almost any circumstance. :)

> Thinking more about this, I believe this can be done without
> adding too many complications.  Proof left as an exercise:-)

With a traditional raid5/6 it shouldn't be too hard doing a
hare/tortoise re-write from beginning to end provided that you have
reserved some bit of space on the underlying disks for being able to do
this in a crash-safe manner - or am I missing something? In practice I
can imagine that it quickly gets complicated of course; especially
supporting it online. If nothing else, the fact that most
implementations don't support this is probably saying something :)

Wonder if raidz/raidz2 will allow striping to vary on a per-stripe
basis. If so, should it perhaps be just a matter of re-writing the
entire tree with the new disk in place? But then perhaps
snapshots/clones will complicate matters.

Come to think of it, I think this was already discussed on zfs-discuss.

-- 
/ Peter Schuller

PGP userID: 0xE9758B7D or 'Peter Schuller <peter.schuller_at_infidyne.com>'
Key retrieval: Send an E-Mail to getpgpkey_at_scode.org
E-Mail: peter.schuller_at_infidyne.com Web: http://www.scode.org



Received on Mon May 14 2007 - 16:50:19 UTC

This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:39:10 UTC