ZFS - whole disk or partition or BSD slice?

[Shane Ambler]

On 28/01/2013 10:27, james wrote:
> I have a 9.1 system with some SATA disks in RAIDZ, upgraded from 9.0.
>
> The disks are all the same type, and I formatted them for FreeBSD and
> put ZFS in a slice covering most of them.
>
> I have seen suggestions for OpenIndiana etc that it is better to let ZFS
> have the whole raw disk and that this can control the way it manages the
> disk writeback mode.
>
> Does this apply to FreeBSD and ZFS too?
>
> Presumably the disks are currently FreeBSD-specific.  If I used raw
> disks instead of slices, could I read them from a Solaris system too?

I recall reading that using partitions for zfs on FreeBSD was as good as 
full disks. For a boot zpool we need to at least have a partition for 
the boot-code and one for zfs preventing the use of a full disk.

ZFS is meant to be compatible between different endian systems (x86 and 
sparc) From what I have read and heard it sounds like zpools are 
expected be compatible between different OS's as well - as far as zpool 
versions are compatible - but I do expect it would depend on the 
partition tables being readable - while full disk usage should work I 
would also think GPT is compatible. OSX 10.5 (x86 and ppc) included a 
read-only zfs kext (before Apple canned the project) so it must have 
been able to read Solaris or FreeBSD created zpools which does indicate 
a fairly high level of compatibility.

I believe the way ZFS marks disks/partitions with the zpool data is so 
that the zpools can be recognised between systems and controllers - it 
would be interesting to know if and under what conditions a zpool can be 
accessed, both between different FreeBSD machines as well as the 
possibility of reading on a Solaris/Indiana machine. Anyone have the 
resources to test?