XFS filesystem revisited

[Collins Richey]

On Mon, 24 Nov 2003 18:15:27 -0500 (EST) Net Llama! <netllama at linux-sxs.org>
wrote:

> On Mon, 24 Nov 2003, Collins Richey wrote:
> > On Fri, 21 Nov 2003 06:12:55 -0500 "David A. Bandel" <david at pananix.com>
> > wrote:
> >
> > > -----BEGIN PGP SIGNED MESSAGE-----
> > > Hash: SHA1
> > >
> > > On Thu, 20 Nov 2003 23:07:28 -0500
> > > "Brett I. Holcomb" <brettholcomb at charter.net> wrote:
> > >
> > > > I think that the Gentoo people really don't have much experience with
> > > > XFS at all -they seem to favor ext3.  Yes, on the log - I've seen that
> > > > when I had to do an xfs_repair and it told me that I needed to mount
> > > > then umount to get the log data updated.  Once I did that xfs_repair
> > > > fixed it.  I never want to see another ext2 (and maybe 3 <G>) system.
> > > >
> > >
> > > I suspect that there are some Gentoo followers (those that wrote the
> > > Gentoo XFS nonsense) that follow the OpenMosix lead developer, who also
> > > seems to have a dislike for XFS.  I've compared patches from XFS and
> > > OpenMosix (which clash badly) and there's not a lot of difference, but
> > > there is one or two minor differences (and they really may be major for
> > > all I know) in the handling of the base disk I/O.
> > >
> > > I've decided to wait for 2.6, when OpenMosix can no longer ignore XFS --
> > > well, they can, they can just not develop OpenMosix for 2.6.
> > >
> >
> > Net summary is that almost everyone (except for one or two gentoo developers
> > and the OpenMosix folks (I, too, am clueless about OM) loves XFS.  When
> > pressed into a corner, the gentoo developer (who is in the once burned,
> > thrice shy category) allowed the following.
> >
> > "The problem is in how XFS caches content and restores metadata but
> > fills content with null (^@ ) whenever it goes down. We had XFS on our main
> > distribution server and it fucked us over deeply and badly due to this, by
> > chewing in most of the data at a point. This was bad enough to warrant the
> > change. Before that XFS was recommended above others for its performance."
> >
> > "The means of achieving their throughput increase is to whack around with
> > the buffecache and the sync code, making it not flush data to disk every
> > 5 seconds or N blocks (as the current, all other FS do)  which -does- in
> > fact add afurther risk to the files in question."
> 
> My opinion of gentoo has hit an all time low.  That bug was stomped out
> well over a year ago.  And in reality, it never was a bug, it was a
> configurable parameter.  So gentoo apparently doesn't read documentation,
> and then poisons their userbase with their ignornace.  lovely.
> 

Will pass along, anonymously, of course.  This is about the only example of bad
advice that I've gotten from gentoo.

-- 
Collins Richey - Denver Area
if you fill your heart with regrets of yesterday and the 
worries of tomorrow, you have no today to be thankful for.