Re: which to use: ext3, JFS, XFS, ReiserFS?

[Joe Brenner <doom@kzsu.stanford.edu>]

Douglas A. Tutty <dtutty@porchlight.ca> wrote:

> Reiserfs = designed by one person who has had some kind of problems (I
> haven't looked into it).  If damage occurs (e.g. unclean shutdown), may
> not be able to fix the damage and loses data.

I've been using resierfs for some time (including on a flaky laptop)
and I've never seen that problem come up.

Comparing ext3 to reiserfs, version 3 (I haven't used 4 yet),
there's one feature of reiserfs that I like quite a bit: the
reported size of a directory has some relationship to how much
the directory contains.

Under ext3, two empty directories and a copy of the /bin
directory look like this:

  /tmp/test:
  total used in directory 40 available 13126340
  drwxr-xr-x  2 doom doom  4096 2008-01-19 18:40 empty_one
  drwxr-xr-x  2 doom doom  4096 2008-01-19 18:40 empty_two
  drwxr-xr-x  2 doom doom  4096 2008-01-19 18:42 bin_copy

Under reiserfs, they look like this:

  /home/doom/End/Dust/Sound/test:
  total used in directory 7 available 909624
  drwxr-xr-x  2 doom doom   48 2008-01-19 18:41 empty_one
  drwxr-xr-x  2 doom doom   48 2008-01-19 18:41 empty_two
  drwxr-xr-x  2 doom doom 2584 2008-01-19 18:43 bin_copy


For desktop use, the performance benefits of one file system or
another is not likely to be perceivable.  However, reiserfs (and
reiserfs alone, as I understand it) is optimized to handle large
number of small files.  Since I use mh for my mail (which stashes
each mail message as an individual file) I feel a little better
useing reiserfs... (For example, currently, the folder I refile
debian-users mail to has over 50000 files in it, and that's not
the biggest one I have.).

As for Hans Reiser's "personal problems": there are programmers
at Namesys working on both reiser 3 and 4 while Reiser is
unavailable.  It's hardly a reason to avoid the relatively mature
reiser 3.

As for the reliability of reiserfs: haven't had any problems with
it myself.  It's hard for me to see how you can sort out
anecdotal evidence on issues like this: file system failures are
rare enough that no one person's experience is worth all that
much (unless you've been administering clusters of hundreds of
machines with a mixture of different file systems...).