[tech] A question of File System

Ian McKellar ian at mckellar.org
Wed Apr 24 04:11:48 WST 2002


On Mon, 2002-04-22 at 23:20, Nick Bannon wrote:
> On Mon, Apr 22, 2002 at 10:29:14AM -0700, Ian McKellar wrote:
> > On Sun, 2002-04-21 at 07:16, David Luyer wrote:
> > > Fast for writes... reiserfs if you have small files which grow, ext2 and
> > > ext3 are also pretty fast with a small number of files per directory.
> > > 
> > The big advantage of ext3 over reiserfs is that its just ext2 +
> > journalling so its a known quantity. The big disadvantage of ext3 vs
> > reiserfs is that its just ext2 + journalling so its slow.
> 
> ext2/ext3's certainly not "slow". Directory operations are slow on
> large directories. ext3 in full-data-journalling mode is slow at
> writes, not that you'd typically run it that way. Both cream Solaris
> UFS and Veritas VxFS, generally speaking.

I'm sure that ext[23] are a lot faster than the commercial alternatives.
If it does anything well, free software does seem to produce fast
filesystems - take a look at the steps Microsoft is going through to try
to stop Samba - putting whacky restrictive anti-GPL clauses in the
license to their spec.
> 
> (VxFS certainly has its uses - it can do online resizing for example)

That sounds nice.
> 
> ReiserFS is very fast at directory operations on large directories (not
> as fast as ext3+htree, though ::-) ) and large numbers of small files.

htree? that sounds interesting - its tree-based directories for ext[23]?
> 
> > Reiserfs also
> > has (at least it will have) some kind of extended attributes system. I
> > don't know if/when that'll actually be used in userland.
> [...]
> 
> That's definitely got some promise. It's kinda irritating at the moment
> because it doesn't have the simple ext2 attributes (immutable bit,
> etc), but there is or has been a patch to provide that. What it ought
> to get at some stage is more generic extended attributes, eg somewhere
> to store ACL's for Samba. The equivalent ext2 support for that (
> http://acl.bestbits.at/ ) sounds like it's going to go into the kernel
> at a leisurely pace. XFS should also be able to do it.

So I have to get off my arse and do that metadata abstraction in
gnome-vfs I was meant to do two years ago...

Ian
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 232 bytes
Desc: This is a digitally signed message part
Url : http://lists.ucc.gu.uwa.edu.au/pipermail/tech/attachments/20020423/1716c761/attachment.pgp


More information about the tech mailing list