[mythtv-users] Should I use LVM on this new raid?

[Bill Williamson]

On 9/23/07, Dan Ritter <dsr-myth at tao.merseine.nu> wrote:
> On Sun, Sep 23, 2007 at 03:17:31PM +1000, Bill Williamson wrote:
> > On 9/23/07, John Drescher <drescherjm at gmail.com> wrote:
> > > > Unless I'm missing something, ZFS doesn't have a compatible license for
> > > > the linux kernel, hence hasn't been ported.  There is FUSE, apparently
> > > > a user mode implementation, but not sure I want to do down that road.
> > > >
> > > I briefly tested it yesterday at work on a raid that gets over 250MB/s
> > > (read test under hdparm) and my reads and writes with the zfs-fuse
> > > driver were around 25MB/s and 5MB/s with compression. I was very
> > > interested in all its features and everything I did worked but with
> > > that performance it is not usable for me. Possibly it does not work
> > > well with non zfs raid or some other reason.
> > >
> >
> > I've not hit performance that horrible.  ZFS is expecting to be in
> > charge of the raw devices, so if you fed it an already-set-up raid
> > array then that was likely your issue.
>
> I thought this was the case when you posted originally, and now
> I'm sure of it.
>
> Bill, you haven't actually used ZFS on a Linux system, have you?
> And definitely not in a Myth box.
>
> Yes, ZFS is great, ZFS is wonderful, ZFS will make toast. But
> only on a Solaris or FreeBSD box right now, and there's doubt
> about how good it is on FreeBSD.

You're half right.

I ran it for a few months on opensolaris to be honest, not on FUSE.  I
had it NFS exported to my mythsystem.  I decommissioned it in a
massive upgrade (got some "real" drives instead of my hordes of old
80-200 gig drives) a while back (hence my email a week or so ago about
reccomended backend archetecture).

So while not "in" a myth box, "for" a myth box.  I loved that I could
get high performance for my myth area while getting high redundancy
for my photos without worrying about what disk was where.

I ran it shortly on linux as a PoC type deal, and didn't notice any
slowdowns on a Myth test, but never ran it fulltime.