[OpenAFS] AFS shortcomings for read/write

Tino Schwarze tino.schwarze@informatik.tu-chemnitz.de
Fri, 24 Oct 2003 09:12:36 +0200


On Thu, Oct 23, 2003 at 02:12:05PM -0700, David Bear wrote:

> > We've had performance and stability issues with readwrite data,
> > however, and strategically we're migrating most readwrite data into
> > NFS/CIFS.  Yeah, this comes at a huge loss in manageability and
> > security, but life's a bunch of trade offs...  The bottom line is that
> > AFS is suboptimal for readwrite data, and with the availability of
> > MP-fast NFSv3 servers and clients, AFS just can't compete when raw
> > performance is an issue.
> 
> This may be related to the overall desing of AFS -- focusing on
> readonly or write-seldom type of file serving.

I think that read-write-replication raises a lot of issues and therefore
it was wise to design AFS to only support RO replicas with "manual"
syncing.

> The ugly wart that hits me with AFS is the lack of file locking which
> has been discussed many times before.  
> 
> I wonder if rather than changing the AFS server portion, we might look
> at extending the AFS client piece to be an intelligent multi-client --
> one that would serve both AFS and connect to NFS or CIFS (my
> preference is SAMBA)

I don't think so. This sounds like an ugly hack - just for supporting
file locking?! To me it looks like file locking is still not important
enough - we'd have support for it already if it was.

Bye, Tino.

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