[OpenAFS] OpenAFS or Coda?

[Tino Schwarze]

On Sat, Dec 07, 2002 at 08:48:03PM -0800, Michael Loftis wrote:

> The delay of writing 100+MB files out across the T-1 wouldn't be 
> acceptable.  The writes occur pretty equally at both sites, with no clear 
> winner on that.

This is a very difficult scenario.

> This is the reason I'm looking at distributed filesystems, is this is 
> truely a need for that, both sides need high-speed RW access to the files.
> 
> So am I right in understanding that AFS doesn't allow multiple RW replicas? 

Yes, you are right.

> You have one 'master' RW, then other RO replicas, and when a write occurs, 
> what, the whole file goes upstream, or only the changes from the clients 
> cache?  

Almost. The synchronisation is always done explicitly by issuing a "vos
release" command either manually or automatically by some self-made
mechanism. The cloning process is pretty efficient, but I'm not sure
whether it would recognise changes inside files and transmit them
efficiently (like for example rsync does).

> I understand the difficulties in having multiple RW replicas, the 
> locking issues, etc, but the concern is to try and keep from transmitting 
> entire 100MB files all the time and to allow both teams to work efficiently.

These issues are the reason that there seems to be no well-performing
distributed file system offering multiple RW replicas (yet?). Intermezzo
might be able to do it but it's pretty experimental as far as I know.

Bye, Tino.

-- 
             * LINUX - Where do you want to be tomorrow? *
                  http://www.tu-chemnitz.de/linux/tag/