xfs
[Top] [All Lists]

Re: [reiserfs-list] Re: benchmarks

To: Seth Mos <knuffie@xxxxxxxxx>
Subject: Re: [reiserfs-list] Re: benchmarks
From: Xuan Baldauf <xuan--reiserfs@xxxxxxxxxxx>
Date: Sat, 14 Jul 2001 18:24:57 +0200
Cc: Xuan Baldauf <xuan--reiserfs@xxxxxxxxxxx>, Federico Sevilla III <jijo@xxxxxxxxxxxxxxxxxxxx>, linux-xfs@xxxxxxxxxxx, reiserfs-list@xxxxxxxxxxx
References: <Pine.BSI.4.10.10107141752080.18419-100000@xs3.xs4all.nl>
Sender: owner-linux-xfs@xxxxxxxxxxx

Seth Mos wrote:

> On Sat, 14 Jul 2001, Xuan Baldauf wrote:
> > Federico Sevilla III wrote:
> > > For your Squid cache, don't go XFS, but instead go ReiserFS. It will do a 
> > > great
> > > job there. For your Samba or NFS partition, go XFS and not ReiserFS.
> >
> > I do not use XFS, currently. Why is XFS better for Samba?
>
> Because word documents average 0.1 - 1MB.

Ah. So you just mean speed (quantity), not some features (quality)-

>
>
> > > Aside from
> > > the fact that ReiserFS is having problems with NFS (and SFS which runs on 
> > > top
> > > of NFS),
> >
> > NFS has problems with ReiserFS (NFS has design bugs which do not seem to 
> > matter on
> > other filesystems).
>
> Dare to say that it was designed for Unix like file systems with inodes?

It was designed for file systems which have directories which never shrink and 
are
searched linearily rather then by other possible indexing methods (hashes, 
trees, etc.).

It was designed for file systems where you can assume that the same file 
metadata has
the same position over unlimited time into the future.
It was designed for file systems where you can assume that the position 
occupied by
metadata of one file may not be occupied by metadata of another file if the one 
file was
deleted and the other file was created, for an unlimited time into the future.

Due to its "stateless" nature, its restrictrions are applied to the whole 
filesystem
rather than to a fancy NFS server which can handle the problems on a 
case-by-case basis.
This is because without state, there is not known when and if a nfs client will 
access
some resource of the filesystem (again), so optimzing for the cases where you 
can assume
that some reference handle (like inode number, directory cookie, etc) once 
given to an
NFS client will not be reused is not possible.

I hope that someday, there will be an emancipation from NFS away (or an NFS 
without
those problems).
Maybe Samba|CIFS can be a replacement someday. CIFS has some overhead, too, due 
to it's
windows origins, but this is per packed basis and not for an unlimited time 
into the
future.

Are there alternatives? Intermezzo does not seem to be developed anymore. Is 
Coda
usable?

Does XFS have dynamically shrinking directories? If yes, how does XFS handles 
limited
32bit NFS cookies?

>
>
> Cheers
> Seth

Xuân.



<Prev in Thread] Current Thread [Next in Thread>