The minimum amount of memory needed to use ZFS.

[Bob Bishop]

Hi,

> On 23 Dec 2015, at 12:14, andrew clarke <mail at ozzmosis.com> wrote:
> 
> On Wed 2015-12-23 21:43:37 UTC+1100, Stephen Hocking (stephen.hocking at gmail.com) wrote:
> 
>> Inspired by this article:
>> http://arstechnica.com/information-technology/2015/12/rsync-net-zfs-replication-to-the-cloud-is-finally-here-and-its-fast/
>> 
>> I am wondering about changing my offsite back strategy, which currently is
>> made up of a Raspberry Pi with an external 3TB drive sitting at my
>> brother's house, with periodic manual rsyncs. I'd like to change that to
>> doing zfs replications.
>> 
>> I want to use some of my ARM based hardware as the target for the ZFS
>> replication, owing to its low power usage. I have a few Cubiboxes floating
>> around with around 2G of RAM, and a RPI2 or a Banana Pi with 1G. It'd have
>> a UFS root on the SD card, and ZFS on the external drive.
>> 
>> Any ideas?
> 
> I'm curious about this too.
> 
> Currently I run a root-on-ZFS FreeBSD 10.2 amd64 system with 2 GB RAM
> that I use for offline backups. The ZFS pool consists of 2 x 1 TB
> drives in a mirror setup. I've never had FreeBSD run out of memory on
> that machine.
> 
> I suggest you avoid using the deduplication feature of ZFS which from
> what I understands likes to chew through memory.
> 
> I don't use ZFS snapshots on that machine, so can't speak about their
> memory usage. Perhaps it's fairly insignificant, though.
> 
> An alternative might be to use something like rsnapshot, still on ZFS.
> 
> You might get a bigger audience if you ask on the freebsd-questions
> list.
> 
> Regards
> Andrew

FWIW we have a backup box currently running 9.2 amd64 with 4GB RAM and a ZFS mirror. We use rsync to transfer the data daily, and ZFS snapshots to maintain a Time Machine-like structure (currently something over 150 snapshots in play). We did have some instances of apparent memory exhaustion until we limited vfs.zfs.arc_max to 2GB; that doesn’t seem to have affected performance.

Deduplication seems like a very bad idea unless you have both a lot of duplicated data and a serious shortage of disk. It needs a lot of RAM, increasing over time. Depending on the hardware and the use case, compression (which effectively only costs CPU) might be a better option.

--
Bob Bishop
rb at gid.co.uk