Skip to Content.
Sympa Menu

sm-discuss - Re: [SM-Discuss] fresh install questions

sm-discuss AT lists.ibiblio.org

Subject: Public SourceMage Discussion List

List archive

Chronological Thread  
  • From: Phil/CERisE/KG6MBQ <cerise AT littlegreenmen.armory.com>
  • To: sm-discuss AT lists.ibiblio.org
  • Subject: Re: [SM-Discuss] fresh install questions
  • Date: Thu, 25 Jul 2002 17:15:37 -0700 (PDT)

Umm, what I suspected was the cause might very well not be. Let's be a
little more fair to the good folks at SGI. All they say on their site
is "If it hurts...DON'T DO IT!!!"

No, I didn't make that up.

After some brief tests, it was clear on my system that XFS was
significantly faster. That's the reason I use it. I think it's an
excellent filesystem. There may or may not be some logical way around
the problem. I never looked into that. Mostly because I never lost
anything important. Perhaps there's a method to that madness? Perhaps
it involves having backups around? ; )

One thing that should be noted for the ultra-paranoid is that deleted
files are often the first place that XFS tries to write, thus making
unremove operations impossible.

-Phil/CERisE

Ryan Abrams said:

> Right. But metadata should be enough. I guess what I am saying is that it
> shouldn?t consider the write action completed (and thus change the inode
> journal entry) until all the data is written. Changing it before hand is
> just stupid... Which is what xfs would have had to do. Otherwise the journal
> would load and have a log of what the old inode was.. Thus no zeroed files.
>
> But even if I don?t know what the hell I am talking about (and its a
> distinct possibility) - a file system that erases the old file pointer on
> open, and THEN writes the new file.. Its um. Dumb.
>
> -Ryan
>
> On 7/25/02 3:41 PM, "Casey Harkins" <charkins AT upl.cs.wisc.edu> wrote:
>
> > IIRC, ext3 is the only of the three that journals data portions of the
> > filesystem. Reiser and XFS only journal the file metadata (name, size,
> > inodes, perms).
> >
> > -casey
> >
> > On Thu, 25 Jul 2002, Ryan Abrams wrote:
> >
> >> That is an absolutely ridiculous way to handle files in a modern file
> >> system. To the point that if it's true, I suggest no one use XFS.
> >>
> >> The whole point of journaling is that it keeps a journal of what it's
> >> doing,
> >> so that interruptions don't break things. So if XFS moves the pointer, it
> >> should have a journal entry saying where it was moved from. If the file
> >> isnt
> >> properly closed, the journal should revert it back to it's old space.
> >> Otherwise the whole advantage is lost.
> >>
> >> Basically, if you have a journaling file system that zeros out the whole
> >> file if it's open when you crash, you need a different file system...
> >> Because that one's on crack.
> >>
> >> Course, it may have been an old bug they fixed. :)
> >>
> >> -Ryan
> >>
> >> On 7/25/02 4:00 AM, "Phil/CERisE/KG6MBQ"
> >> <cerise AT littlegreenmen.armory.com>
> >> wrote:
> >>
> >>> bluebird AT optushome.com.au said:
> >>>
> >>>> That's interesting, I thought that a major advantage of
> >>>> journaling file systems like Reiser & XFS was there ability
> >>>> to recover from sudden crashes.
> >>>>
> >>>> Cheers
> >>>>
> >>>> Joel
> >>>
> >>> Well, the problem in the case of XFS is with open files. I don't
> >>> know the internals well enough to comment on why they're suddenly
> >>> replaced with binary zeros. I suspect it's because when it opens a
> >>> file, it preallocates a space for it and sets all the appropriate
> >>> pointers to the new place on disk.
> >>> They recover faster in the sense that fsck has very little work to
> >>> do by comparison. They don't necessarily recover better.
> >>> With this in mind, I haven't had that problem with XFS for a good
> >>> long while. About 5 months or so of constant use have gone by without
> >>> problems.
> >>>
> >>> -Phil/CERisE
> >>>
> >>> _______________________________________________
> >>> SM-Discuss mailing list
> >>> SM-Discuss AT lists.ibiblio.org
> >>> http://lists.ibiblio.org/mailman/listinfo/sm-discuss
> >>
> >> _______________________________________________
> >> SM-Discuss mailing list
> >> SM-Discuss AT lists.ibiblio.org
> >> http://lists.ibiblio.org/mailman/listinfo/sm-discuss
> >>
> >
>
>





Archive powered by MHonArc 2.6.24.

Top of Page