[Discuss] ZFS for an Encrypted External Disk: My Experience and Notes

[richard.pieri at gmail.com (Rich Pieri)]

On Tue, 26 Jul 2022 14:41:55 -0700
Kent Borg <kentborg at borg.org> wrote:

> It took hours. The fact that ZFS assumes that numerically distant
> LBAs will be physically distant, to the extent that assumption holds,
> will make for a lot of extra head movement. Maybe Linux can reorder
> writes around that a little, but it is a pessimizing thing to do for
> speed. (Issuing parallel writes to two different drives suddenly
> seems like such a luxury.)

If you are getting ~100MB/s sustained then that is about as fast as you
can go on spinning disks regardless of the bus. If you're getting less
then the likely culprit is a wrong block size. ZFS defaults to 512B
blocks but if you have 4K disks then you want to specify a 4K ashift
when creating the pool:

  zpool create tank -o ashift=12 [...]

For those considering SSDs with 8K blocks you want ashift=13.

> Were the disk I'm backing up also ZFS, it could be done much more
> efficiently and quickly and precisely, but by doing it this old
> fashioned rsync way I/do/  get the side effect of a partial, kinda
> half-scrub, specifically of the data that is of current interest.

Yes. zfs send/receive bypasses all the overhead of walking file
systems, and transfers snapshot deltas directly from source to
destination. Much faster than rsync but it is more complex because you
have to manage snapshots.

-- 
\m/ (--) \m/