Jim Mostek wrote:
>
> I haven't played with ReiserFS but from what I've read/heard, my guess
> is that it will do small files really well. XFS' directories can
> also handle many entries, but ReiserFS with its packing and data in the
> inode can handle small files better. But, I don't know how ReiserFS will
> do when many processes are hitting the same directory on a multi-CPU
> system. XFS has had lots of work done in this area since we have many
> customer running old sendmail which does all its file work in the same
> directory.
We are having a friendly respectful little race to see whether XFS can port
faster than we can tune. The XFS guys are extremely skilled programmers, but
they have a large code base, and Network Appliance hired many of the writers of
that code. I think that ReiserFS's core tree technology is ahead of XFS, and we
are working on a complete rewrite which will deploy new next generation tree
technology (we are leaving our B* trees behind for something much more
powerful), meaning we hope to see the gap widen a bit while they focus on
porting rather than rewriting. We just had a two day seminar on the new
algorithms, so I am a bit jazzed about them.:) That said, I would be very
surprised if there were no areas where XFS was better than ReiserFS when XFS
comes out, the XFS guys are good....
It may amuse you to know that both sides have tried to convince the other team
to quit and join their effort, and this is an indication of our mutual respect.
> XFS will have additional capabilities that the other file system you
> mentioned don't have such as Direct I/O and extended attributes.
If by extended attributes you mean variable allocation of stat data, we are
doing that, but it is not shipping yet. We are also doing I/O scheduler work
that is still very raw but may be usable by summer. I would hope that there is
code the XFS team puts into VFS that we can use and vice-versa.
> At least
> I haven't heard that others are adding these.
>
> Longer term, it appears that the Linux community will have a rich set
> of file systems available. These can be downloaded and analyzed.
> I suspect that there will be some matrix mapping applications/environments
> to file systems to show which file system(s) to use for each.
>
> Jim
>
> >
> >Hi!
> >
> >We now have several journaling file systems, like XFS, ReiserFS and
> >ext3 (and more I believe, but I would like to concentrate on these
> >three).
> >I have played with ext3 a few weeks ago, which was oke (easy to install
> >and if I unplugged the machine, it was up and running (fine) in no time
> >just the way sysadmins like it. :-)
> >
> >Now it seemes to me that ReiserFS has a larger developer base and aims
> >a bit higher than ext3. Is that true? If not, what is the difference
> >between ext3 and ReiserFS?
> >ReiserFS seemes to be a good choice for spool directories because it
> >can handle a _huge_ amount of files (and dirs) in a dir efficiently.
> >
> >And there is XFS. What is the advantage or disadvantage of XFS compared
> >to ext3 and/or ReiserFS? It seemes to me that XFS is a bit less stable
> >than the others at this moment. Is that true? XFS exists for some time
> >now. Is that an advantage above ReiserFS because it had more time to
> >develop itself to what is is now, or is ReiserFS 'better' because it
> >doesn't have to carry its history and doesn't have to be backwards
> >compatible?
> >
> >Please enlighten me. :-)
> >
> > Ookhoi
> >
|