ZFS questions

[Ronald Klop]

On Wed, 08 Aug 2012 06:24:14 +0200, Don Lewis <truckman at freebsd.org> wrote:

> I've got a couple of questions about a raidz array that I'm putting
> together.  Capacity is more important to me than speed, but I don't want
> to do anything too stupid.
>
> The fine manual says that using whole disks is preferable to using
> slices because drive write-caching is enabled if the entire drive is
> dedicated to ZFS, which would break things if the drive also contained a
> UFS slice.  Does this really matter much if NCQ is available?  Each
> drive will be dedicated to ZFS, but I'm planning on using GPT to
> slightly undersize the the ZFS slice on each drive to avoid any
> potential issues of installing replacement drives that are slightly
> smaller than the original drives.

Solaris does/did this. FreeBSD does not disable the write-cache if you  
don't use the whole disk.

> I'm slowly accumulating the drives over time for both budgetary reasons
> and to also try to reduce the chances of multiple near-simultaneous
> failures of drives from the same manufacturing batch.  I'd like to get
> the array up and running before I have all the drives, but unfortunately
> ZFS doesn't allow new drives to be added to an existing raidz vdev to
> increase its capacity.  I do have some smaller drives and I was thinking
> about pairing those up with gconcat or gstripe and configuring the ZFS
> pool with the concatenated/striped pairs.  I know this isn't
> recommended, but it seems to me like zpool create would accept this.
> What concerns me is happens on reboot when ZFS goes searching for all of
> the components of its pool.  Will it stumble across its metadata on the
> first of the two concatenated pairs and try to add that individual drive
> to the pool instead of the pair?

I don't know. Somebody else might answer this.

Ronald.