FreeBSD + ZFS on a production server?

[Wojciech Puchar]

>
> ZFS is very nice, but slightly over-hyped imho.

not slightly and not only over-hyped. it's definitely far from being for 
storage as "VM is for memory".
for example you can't select per file (or at least - per 
pseudo-filesystem) if you want no protection, mirrored or raidz.

you must have disks dedicated for raidz, disks dedicated for mirrored 
storage and disks dedicated for unprotected storage. it's inflexible and 
not much usable.

actually - much less usable than "legacy" 
gmirror/gstripe/gconcat+bsdlabel.


one of my systems have 8 disks. 80% of data doesn't need any protection, 
it's just a need for a lot of space, other 20 needs to be mirrored.
this 80% of data is used in high bandwidth/low seeks style (only big 
files).

i simply partitioned every disk on 2 partitions, every first is used to 
make gmirror+gstripe device, every second is used to make gconcat device, 
and i have what i need WITH BALANCED LOAD.

with ZFS i would have to make first 2 drives as mirror, another 6 for 
unprotected storage, having LOTS of seeks on first 2 drives and very 
little seeks on other 6 drives. the system would be unable to support the 
load.



to say more: zfs set copies could be usable to selectively mirror given 
data while not mirroring other (using unprotected storage for ZFS).
but it's broken. it writes N copies under write, but don't remake copies 
in case of failure!