Re: [PLUG] Vmware oops

[Rich Freeman <r-plug@thefreemanclan.net>]

On Fri, Dec 9, 2011 at 12:13 AM, Lee H. Marzke <lee@marzke.net> wrote:
> As in my recent Plug ZFS talk ,  I discussed  a known issue with RAID-5 called the
> 'write-hole' \1.   So if you lose power to the array you can wind up with
> silent data corruption where the array still provide correct data, but
> the parity is scrambled.    Now when you actually lose a drive and
> reconstruct the array,  the rebuild silently replaces your data with junk.

That is limited to a single stripe, but with any non-COW filesystem
(or layer below the filesystem) data is being overwritten in place,
and there is the potential for some data loss if that is interrupted.
RAID 5/6/etc can make that worse since typically you have to modify an
entire strip at one time so loss is not necessarily limited to just
files that are open.

This is definitely a limitation on linux unless you're willing to use
ZFS on FUSE (which is only somewhat mature - the code it started with
is mature but the implementation is not), the still-experimental
btrfs, or do your data storage on some other OS that has more mature
COW support.

Long-term the expectation is that the default filesystem on Linux will
become btrfs - which is COW and checksums everything.  The problem
right now is that it is pretty immature and the versions in stable
kernels still panic fairly easily.  Progress seems to have picked up
and my current guess is that adventurous types might start using it
for real data a year from now, and less adventurous types maybe 3
years from now - assuming maturity continues to increase steadily.

Btrfs has a potential feature set similar to ZFS.  In terms of
implemented features there are a few things it does that I don't think
ZFS does yet, but for each of them there are probably 10 things that
ZFS is ahead on.  Its underlying architecture is a little more
advanced which I think gives it a little more potential long-term, but
I think in practice both will be reasonably comparable for a while.
The key difference between them is their license (GPL-compatible vs
not) - this is basically a design feature in both cases.

I think the real irony, however, is that right now the same company is
the driving force behind both of them (Oracle).  I'm not sure what
limitations Oracle is under but it seems odd that they are selling a
non-GPL-compatible COW filesystem and at the same time independently
trying to create a GPL-compatible COW filesystem from scratch.
Interest in Btrfs has of course grown beyond Oracle and they don't own
the btrfs code - so unless it becomes truly obsolete I doubt it is
going away anytime soon (and some might continue to develop it as a
B+-tree alternative to ZFS).

Rich
___________________________________________________________________________
Philadelphia Linux Users Group         --        http://www.phillylinux.org
Announcements - http://lists.phillylinux.org/mailman/listinfo/plug-announce
General Discussion  --   http://lists.phillylinux.org/mailman/listinfo/plug