ZFS guidelines - preparing for future storage expansion
James R. Van Artsdalen
james-freebsd-fs2 at jrv.org
Mon Nov 30 21:08:37 UTC 2009
Zaphod Beeblebrox wrote:
> I moved from 5x 750G to 5x 1.5T disks this way earlier this year.
> [...] And keep in mind that while you're upgrading, you're vulnerable
> to data loss (no more replicas).
This is one of the (many) reasons I prefer mirrors rather than parity
(RAID-5). You can "attach" the new drive, wait for the resilver to
complete, then detach the old drive - never having fewer than two drives
in the mirror. And of course you can gain space in the pool as each
mirror is upgraded whereas a parity group (RAIDZ) usually involves more
drives.
Note that the zpool(1) man page says of the "Replace" command:
"Replaces old_device with new_device. This is equivalent to attaching
new_device, waiting for it to resilver, and then detaching
old_device". This is not quite true: the reads for the resilver come
from all available devices if you do attach/detach, but do not come from
old_device if you do "replace". This is for MIRRORs; I'm not sure how
RAIDZ behaves.
More information about the freebsd-fs
mailing list