Re: alternative file systems
Hi.
On Fri, 17 Oct 2014 03:00:26 +0200
lee <lee@yagibdah.de> wrote:
> Reco <recoverym4n@gmail.com> writes:
>
> > On Sun, Oct 12, 2014 at 03:33:15AM +0200, lee wrote:
> >> > A correct guess. A recommended minimum is kernel 3.14 - [2].
> >>
> >> So this is a rather new feature. How reliable and how well does it
> >> work?
> >
> > I wouldn't trust my data to that feature :) It has 'experimental' and
> > 'biohazard' labels strapped everywhere.
> > I prefer trusty mdadm for any RAID.
>
> One of the disadvantages with mdadm is that it can severely impact
> performance.
Agreed. Still, I view RAID as a disaster prevention tool first, and any
performance increases come only second if they do at all.
> That doesn't mean that raid-5 with btrfs wouldn't have
> this disadvantage, too.
Sure. I'd only wait two or three years before trying it. btrfs by
itself is interesting, it only needs to get rid of those 'experimental'
labels IMO.
> >> > But, ZFS won't allow you to make a conventional RAID5 either :)
> >>
> >> I know --- and I don't require RAID-5. What I require is what RAID-5
> >> provides, i. e. redundancy without wasting as many disks as other RAID
> >> levels. I also like the better performance of hardware RAID compared to
> >> software RAID. IIRC, ZFS would provide efficient redundancy and be
> >> safer than a RAID controller because of it's checksumming. I'd have to
> >> try it out to see what kind of performance degradation or gain it would
> >> bring about.
> >
> > A real story. A recent one, a couple of weeks fresh.
> > One shop buys *very* expensive Sun SuperCluster T4 with Solaris 11 and,
> > of course, ZFS. Configures a couple of LDOMs on it. So far, so good.
> > And then - it happens. A simple oversight - they filled up to 100% one
> > of LDOMs' root zpool.
> > They say that is should not happen, yet I've seen it with my own eyes -
> > ZFS happily ate (i.e. they disappeared without a trace) a couple of
> > shared libraries, rendering some basic OS utilities unusable.
> > So, what good was those magical ZFS checksums did?
>
> And not having the checksumming has never caused a problem for me, as
> far as I can tell ... Still that doesn't mean that it hasn't.
The morale of the story is that checksums are not a silver bullet.
> >> >> They need to get these license issues fixed ...
> >> >
> >> > Back in the old days CDDL was chosen by Sun especially so that
> >> > this license issue would *never* be fixed.
> >> > Currently Oracle could re-license ZFS to anything they want, including
> >> > GPL-compatible license, but why would *they* do it?
> >>
> >> Why don't they?
> >
> > Simple - they sell servers based on Solaris as storage appliances (and
> > they nearly 10 years behind ZFS on Linux as far as ZFS is concerned). Who
> > will buy these servers if the same can be achieved with cheap Linux
> > server? Oracle is greedy.
>
> But when it eats files and is 10 years behind, why are people buying it?
Beats me. Either they use 'more expensive is better' approach, or they
use human beings to watch very carefully that their filesystems do not
overflow.
And of course, such people backup their data usually :)
> So how can we safely store large amounts of data?
As far as long-term storage goes - I prefer LTO7.
As for the short-term storage - I prefer ext4, lvm, mdadm *and* a
backup.
Reco
Reply to: