[OpenIndiana-discuss] About zpools and usage plan

[Timothy Coalson]

On Sat, Aug 3, 2013 at 3:58 AM, Harry Putnam <reader at newsguy.com> wrote:

> "DormitionSkete at hotmail.com" <dormitionskete at hotmail.com> writes:
>
> > On Aug 2, 2013, at 9:08 PM, Harry Putnam wrote:
> >
> >> Looking for a little advice about setting up a home lan server.
> >>
> >> So far just tinkering with a vbox oi as guest install hoping to make
> >> whatever nasty blunders on disposable data and OS.
> >>
> >> But getting right to it:
> >>
> >> In a previous life I ran oi and used mirrored sets of discs for
> >> everthing.
> >>
> >> Mirrored OS discs, and 2 sets of mirrored data discs.
> >>
> >> I was advised back then that mirrored sets of discs, given  my small
> >> usage, would be the most secure (not meaning agains penitration or the
> >> like...just keeping data from loss) and handiest way to handle my home
> >> lan.
> >>
> >> So, wondering what the current thinking is?  Like say, mirrored as
> >> opposed to a collection of discs in raidz or whatever its called.
> >>
> >> My needs will be well handled by something like 3tb of storage so
> >> something like 6 1tb discs for a mirrored setup.
> >>
> >> What else could I get with those 6 disks in terms of redundancy and ease
> >> of maintenance?
>
> They're a little weak in the snapshot area, and throw big sighs when
> asked to scrub a disk.... hehe.
>
> My question here was about the various ways of using a zfs box.
>

Richard Elling did some comparisons of vdev layouts, calculating mean time
to data loss:

http://blog.richardelling.com/2010/02/zfs-data-protection-comparison.html

https://blogs.oracle.com/relling/entry/a_story_of_two_mttdl
https://blogs.oracle.com/relling/entry/raid_recommendations_space_vs_mttdl

The pictures on the oracle blogs got botched, though they can be dug out
(they reference the old sun url, but they did get moved to the new url).
The short version is that while mirror pairs are safer than raidz1 for the
same number of disks, raidz2 is safer than either (and more space efficient
to boot - what you lose is some performance in random workloads).

Tim