Because XFS is BETTER (Re: opinion on XFS)

[Chris Black]

On Thu, May 09, 2002 at 07:39:26PM +0300, Eray Ozkural wrote:
> On Thursday 09 May 2002 17:48, Chris Black wrote:
> > > The only filesystem on linux that would come close to XFS is JFS and that
> > > is said to be still not-so-stable. Forget reiserfs, the docs scream
> > > no-no.
> >
> > Could you elaborate on this? We've been using reiserfs pretty heavily
> > without problems. The main reason to use a non-ext2 fs on a cluster
> > system as I see it is handling directories with many files. Reiserfs
> > does a much better job of this than ext2 in our experience.
> 
> I don't know how reiserfs works in practice, but from the docs it doesn't 
> sound like they have decided on a particular algorithm / structure for the 
> filesystem. The most technical detail I had found was that it used "balanced 
> trees". As you know, that's not a file/data structure.
> 
> I won't use any index structure that doesn't have well established bounds on 
> its time/space complexity.

The only changes I have seen are drop in replacements of the hash table 
code used to handle directory/file lookups. All of them have worked 
for me, and they haven't changed it in quite awhile. In fact in 
reiserfs 3.6.x (the 2.4 kernel version) I don't think they have changed 
the hash algorithm. On 2.2 kernels it changed from tea to rupasov and 
then finally r5. Other changes with the 3.6 version have to do with 
supporting large files, and you can easily change between 3.5 and 3.6 
on-disk formats by mounting a volume with the -o conv option.
I'm sure I'm glossing over some things as I'm not actively involved in 
reiserfs development.

I am generally very happy with reiserfs (especially under 2.4), in
particular its handling of directories with thousands of results files
from cluster jobs.

Chris

-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 232 bytes
Desc: not available
URL: <http://www.beowulf.org/pipermail/beowulf/attachments/20020509/c41fce06/attachment.sig>