Re: UFS filesystem size limit

[Joerg Sonnenberger <joerg@xxxxxxxxxxxxxxxxx>]

On Sat, Sep 03, 2005 at 04:13:52PM -0000, Andreas Hauser wrote:
> joerg wrote @ Sat, 3 Sep 2005 14:09:39 +0200:
> > Don't compare experimental filesystems with UFS :-)
> 
> While not having as much testing behind them as UFS, at least on Linux,
> they aren't experimental filesystems anymore.

I was explicitly refering to ReiserFS. Ext2/Ext3 are stable to a certain
degree, as is JFS. I'd trust XFS the most on Linux.

> > But seriously, the
> > structure of JFS and XFS is very different from UFS, e.g. the use of
> > btrees for almost anything. That makes them more suitable for some
> > operations, but horrible for others.
> 
> Can you name the operations where those are less suitable?
> Keep in mind they don't use plain B-Trees.

The question is whether you have direct access in UFS or not. For a
large directory, opening a specific file is faster with on-disk b-trees,
but e.g. a full directory read is not faster with b-trees, it might even
be slower due to increased overhead.

Similiar arguments apply to extend based storage like XFS does for large
files, it does make some operations slower and some faster. It is always
a balance.

> It might very well be that one of the Linux FS is better in all cases
> than UFS.

Sure. It should also make you a bit nervous because it is a tradeoff
between simple on-disk representation and speed :-) The chances to
recover UFS after e.g. a partial disk failures are higher than most
other filesystems, because the simple structure is more resistent to
failures. That's also something to keep in mind.

Joerg