Hi,

I have read in last couple of days more about ZFS then I believe is possible
(I like http://www.opensolaris.org/os/community/zfs/demos/ -- the
term "geek porn" comes to mind ;-)), but there is still one thing which I
don't understand that well. ZFS is claimed to be "self-healing"
<http://www.tech-recipes.com/solaris...n_tips1446.html>
even without HW RAID. Does this self-healing means just that ZFS has
equivalent of SW RAID inside (RAID 1 and RAID 5 aka RAID-Z seem to be
available), or how else it could restore damaged data? Is "self-healing"
just marketing hype, or they found other way how skin the cat?

This is probably the most silly question, but bear with me please I would be
really like to be sure about the answer.

Thanks in advance,

Matěj Cepl
--
GPG Finger: 89EF 4BC6 288A BF43 1BAB  25C3 E09F EF25 D964 84AC
http://www.ceplovi.cz/matej/blog/, Jabber: ceplma@jabber.cz
23 Marion St. #3, (617) 876-1259, ICQ 132822213

Faithful love is what people look for in a person; ...
-- Proverbs 19:22 (NJB)

[ Post a follow-up to this message ]

Matej Cepl wrote:
> Hi,
>
> I have read in last couple of days more about ZFS then I believe is possib
le
> (I like http://www.opensolaris.org/os/community/zfs/demos/ -- the
> term "geek porn" comes to mind ;-)), but there is still one thing which I
> don't understand that well. ZFS is claimed to be "self-healing"
> <http://www.tech-recipes.com/solaris...n_tips1446.html>
> even without HW RAID. Does this self-healing means just that ZFS has
> equivalent of SW RAID inside (RAID 1 and RAID 5 aka RAID-Z seem to be
> available)

No, since conventional SW and HW RAID are not self-healing - they merely
provide for continued operation in the event of a single disk failure,
without necessarily restoring the earlier level of redundancy.

ZFS, by contrast, not only provides its software RAID facilities but
also attempts to use them to restore the original level of redundancy
after a disk failure if sufficient distributed free space exists within
the 'storage pool' (conceptually like using that distributed free space
to fulfill the role of one or more 'hot spare' disks in more
conventional systems, but arguably more flexible and extensible).  I
don't believe this is a unique feature (e.g., HP's EVA storage boxes and
IBM's Storage Tank may do something similar at the block level), but it
is at least currently an unusual one, especially in a file system
available to mere mortals.

- bill

[ Post a follow-up to this message ]

Bill Todd wrote:
> ZFS, by contrast, not only provides its software RAID facilities but
> also attempts to use them to restore the original level of redundancy
> after a disk failure if sufficient distributed free space exists within

Thanks.

Matěj

--
GPG Finger: 89EF 4BC6 288A BF43 1BAB  25C3 E09F EF25 D964 84AC
http://www.ceplovi.cz/matej/blog/, Jabber: ceplma@jabber.cz
23 Marion St. #3, (617) 876-1259, ICQ 132822213

As a rule of thumb, the more qualifiers there are before the name
of a country, the more corrupt the rulers. A country called The
Socialist People's Democratic Republic of X is probably the last
place in the world you'd want to live.
-- Paul Graham discussing (not only) Nigerian spam
(http://www.paulgraham.com/spam.html)

[ Post a follow-up to this message ]

Bill Todd <billtodd@metrocast.net> wrote:
> No, since conventional SW and HW RAID are not self-healing - they merely
> provide for continued operation in the event of a single disk failure,
> without necessarily restoring the earlier level of redundancy.

Many/most existing SW and HW RAID are "self-healing" to some extent -
although none to the extent that ZFS is.

When they detect a bad block (in particular, a read error), most RAID
systems will re-create the data (from the mirror or parity) and re-write
the bad block. This will generally result in a reallocation occuring on
the disk, and the region where the error on the disk occured being mapped
out.  Unfortunately SVM is one of the ones that doesn't do this, although
VxVM does.

You could also argue that hot sparing is also a form of self-healing -
when a disk fails, the VM will automatically "heal" the fault by
replacing the failed disk with the hot spare.

ZFS takes this much further with things like end-to-end checksumming.
There's heaps of details on this on the OpenSolaris ZFS website -
http://opensolaris.org/os/community/zfs/

Scott.

[ Post a follow-up to this message ]

Scott Howard wrote:
> Bill Todd <billtodd@metrocast.net> wrote: 
>
> Many/most existing SW and HW RAID are "self-healing" to some extent -
> although none to the extent that ZFS is.
>
> When they detect a bad block (in particular, a read error), most RAID
> systems will re-create the data (from the mirror or parity) and re-write
> the bad block. This will generally result in a reallocation occuring on
> the disk, and the region where the error on the disk occured being mapped
> out.  Unfortunately SVM is one of the ones that doesn't do this, although
> VxVM does.
>
> You could also argue that hot sparing is also a form of self-healing -
> when a disk fails, the VM will automatically "heal" the fault by
> replacing the failed disk with the hot spare.

That's all well and good, but it's not part of the conventional
definition of RAID even if many implementations may augment that
definition by including such features.  That's why I used the word
'necessarily' in my original statement above.

By contrast, self-healing *is* part of the definition of ZFS.

- bill

[ Post a follow-up to this message ]