[geeks] Solaris 10 / OpenSolaris bits to be in next version of OSX

[Charles Shannon Hendrix]

Thu, 10 Aug 2006 @ 10:12 -0400, der Mouse said:

> You don't, actually.  If something is failing to write data, you don't
> (can't) know whether it's the data or the saved checksum that's been
> mis-written, and have no particular reason to trust the redundant copy
> either.  It too could be out of date, after all; if the disk (or disk
> subsystem software) can lose one write, it can lose two writes.

ZFS, as described here, seems to be falling into the "who will guard the
guardians" trap in reliable systems software.

It also seems that ZFS is a lot newer and untested than most of the
firmware in drives.

Contrary to what has been said here, undetected write failures on good
modern drives is fairly rare.  If it really bothers you, you can buy
drives that do write-verify cycles which even further reduce undetected
errors.

What I'm leading up to is this: what happens when a drive says the write
was OK, but ZFS says it wasn't?

In most cases, given the newness of ZFS code, I'd be worried about
letting ZFS override what the drive thinks.

So... how does ZFS handle disagreements between itself and storage
hardware?  Does it always trust itself?




-- 
shannon "AT" widomaker.com -- ["Meddle not in the affairs of Wizards, for
thou art crunchy, and taste good with ketchup." -- unknown]