[Dclug] Slides up, was Re: DCLUG meeting reminder: tonight, Wed, Aug 19, 7pm: BTRFS

[Jesse Becker]

On Thu, Aug 20, 2009 at 16:50, Przemek
Klosowski<przemek at jazz.ncnr.nist.gov> wrote:
>   A minor request though:  On slide 8, one of the items reads "XFS from
>   SGI, which is on the ropes."  Could you elaborate a bit on this?  Do
>   you mean that SGI is "on the ropes" (which is not news), or XFS is "on
>   the ropes" (which is news).  :)
>
> I hope you are not going to base your strategic filesystem deployment
> decisions on jokes from my presentation :)

You mean, that's not a good way to deploy critical infrastructure
components?  Darn.

Actually, at $DAY_JOB, we're heavily invested in XFS, with about 100TB
of storage in multiple XFS filesystems.  EXT3 doesn't support
sufficiently large filesystems in the mulit-TB range.  The limit is 8
or 16TB--I forget which--but neither is large enough.  JFS really does
seem stagnant, in large part, and I've had mediocre luck with it in
the past.  ReiserFS[34] aren't options either since they are no longer
really supported either.  EXT4 wasn't nearly as close to stable at the
time of deployment, and BTRFS is still a ways off as well.  So XFS was
really the only viable option at the time.  Now, I'd consider both XFS
and EXT4.

> I meant to say that SGI was in a tight spot, even back in 2007 when
> XFS was certainly one of the best filesystems around. Two years later,
> with SGI bought by Rackable, in all seriousness, I would worry about
> XFS possibly succumbing to bit rot.  It doesn't _have_ to happen, and
> XFS surely works well today, but if it is not developed actively, I
> could see it having trouble keeping up with evolving kernel
> infrastructue (VFS and whatnot).

Agreed on all counts, and I'll hold up JFS as a potential example.
While I'm sure that JFS "works" (surely we'd have heard about it on /.
if it didn't), there appears to be almost zero active development.
XFS on the other hand, seems to be still alive in many respects.  I
think that there were several commits to the mainline kernel recently
for inclusion in 2.6.30 or .31.  So for the moment, XFS appears to be
"alive."  We'll see what happens next year, and the year after that...

> BTW, it's the same reason why in my personal opinion BTRFS will
> fare better than ZFS, even if today it is quite raw(*): there are
> more people banging on it. Self-fullfilling hypothesis, I know.

That might depend on your definition of "fare better".  In terms of
number of filesystems, I think in the long term you're probably
correct.  ZFS is currently tied to Solaris as mentioned before.  This
severely restricts the number of potential systems that will could use
it.  BTRFS could eventually make it into many/most of the new Linux
installs that will be seen a few years hence, and could have a very
high filesystem count.  (However, I recall a time when people thought
reiserfs3 had a good chance of being very widely used...)

That said, ext3 is probably the most common filesystem on non-embedded
servers, but it "fares poorly" (IMO) on techinical grounds relative to
many of the other filesystems (XFS, BTRFS, ZFS, JFS, ReiserFS[34]) in
terms of features and performance.  Similarly, on Solaris, Veritas
is/was a technically superior filesystem to UFS, but it doesn't have
nearly the same install base (it is commercial software, of
course...).  I wouldn't quite say that Veritas fared "poorly", but I'm
not sure that it "fared well" either. :-)


-- 
Jesse Becker
GPG Fingerprint -- BD00 7AA4 4483 AFCC 82D0  2720 0083 0931 9A2B 06A2