Filesystem wars! (was Re: woody CDs :()

[Kevin Buettner]

On Mar 27,  1:18pm, Blake Barnett wrote:

> On Wed, 2002-03-27 at 11:58, Kevin Buettner wrote:
> > I have one ReiserFS partition, but I don't use it much.  The thing
> > that scares me about ReiserFS is that I've heard there are (were?) NFS
> > compatibility problems.  Also, I've heard tales of filesystem corruption
> > with ReiserFS.  Perhaps these problems are fixed now; I haven't checked
> > recently.
> > 
> I've seen reports even recently of some pretty harsh corruption with
> Reiser, I've also seen reports of systems running without problems for
> long periods of time with HUGE amounts of data... as always YMMV.

Okay, thanks for the update.

> > I'm still using ext2 and ext3 for most of my filesystems.
> 
> Stick with them, if they do everything you need, why change?

That's been largely my attitude.  However ReiserFS has advantages
beyond mere journaling, and it doesn't hurt to experiment with
non-critical data.

> > > BTW if anybody is using Coda successfully I could use help with that too.
> > 
> > I played around with Coda recently, but there was a pretty severe bug
> > that prevented me from going very far with it.  It looks to me like
> > this is a dead end project; I think the Coda developers have switched
> > to Intermezzo.
> 
> Yes, coda is almost purely a research project, not really worth trying
> to use for anything near production.
> > 
> > > Meanwhile Intermezzo seems too immature and only works on top of an ext3
> > > filesystem.
> > 
> > I concur.
> 
> Intermezzo is the result of the research on Coda... Give it a little
> more time and it'll be pretty usable.  But as always, you have to ask
> yourself if you really need disconnected operation.  The benefits may
> not be as great as you think...  Mr. Braam (the primary author) said
> these very words to me.   NFSv4 may be a more viable solution for most
> problems.

The problem that caused me to look at Coda and Intermezzo recently is
that I have a machine for which NFS is not very stable, i.e, the machine
usually crashes shortly after doing an NFS mount.  For reasons that I
won't go into, I'm not permitted to change to a (potentially) more
stable kernel.  It seemed to me that perhaps Coda or Intermezzo would
work where NFS would not.

I ended up not using either of these solutions.  All I really care
about is pushing the contents of my development directory over to
the (unstable) machine as it changes over time.  I ended up writing
a little script that interfaces with fam (the File Alteration Monitor)
that ships with recent Red Hat releases.  The script uses the services
provided by fam to notice when a file has changed and then invokes
rsync over ssh to update it on the remote machine.  It's smart about
recognizing when a number of files change within a short span of time
(e.g. during a cvs update) and syncs these files in a batch rather than
individually.

The nice thing about this approach is that I don't need privileged
(root) access to either the local machine or the remote machine to use
it.  The only problem with it is that the implementation of fam that
ships with Red Hat Linux uses polling to determine when the files /
directories that it's monitoring has changed.  If you're monitoring a
number of large directory trees, it can be somewhat CPU intensive. 
SGI has a kernel patch which allows fam to be used without polling,
but I haven't tried it yet.

Kevin