New ZFSv28 patchset for 8-STABLE: Kernel Panic

[Jean-Yves Avenard]

On Wednesday, 29 December 2010, jhell <jhell at dataix.net> wrote:

>
> Another note too, I think I read that you mentioned using the L2ARC and
> slog device on the same disk.... You simply shouldn't do this it could
> be contributing to the real cause and there is absolutely no gain in
> either sanity or performance and you will end up bottle-necking your system.
>
>>

And why would that be?

I've read so many conflictinginformation on the matter over the past
few days that I'm starting to wonder if there's an actual definitive
answer on the matter or if anyone has a clue regarding what they're
talking about.

It ranges from, should only use raw disk to freebsd isn't solaris so
slices are fine. Don't use slice because they can't be read by another
OS use partitions..
It doesn't apply to SSD and so on..

The way I look at it, the only thing that would bottleneck access to
that SSD drive, is the SATA interface itself. So using two drives, or
two partitions on the same drive, I can't see how it would make much
difference if any other than the traditional "I think I know"
argument. Surely latency as with know it with hard drive do not apply
to SSDs.

Even within sun's official documentation, they are contradicting
information, starting from the commands on how to add remove/cache of
log device.

It seems to me that tuning ZFS is very much like black magic, everyone
has their own idea about what to do, and not once did I get to read
conclusive evidence about what is best or find an information people
actually agree on.

As for using unofficial code, sure I accept that risk now. I made a
conscious decision on using it, there's now no way to go back and I
accept that.
At the end of the day, it's the only thing that will make that code
suitable for real world condition: testing. If that particular code
isn't put under any actual stress how else are you going to know if
its good or not.

I don't really like reading between the lines of your post that I
shouldn't be surprised should anything break or that it doesn't matter
if it crashes. there's a deadlock occurring somewhere : it needs to be
found. I know nothing about the ZFS code, and I could only do what I'm
capable of under those circumstances: find a way to reproduce the
problem consistently, report as much information as I have so someone
more clueey will know what to do with it.

Hope that makes sense

Jean-Yves