On 12/09/07 03:44, Michael Semcheski wrote:
> This is a question regarding OpenAFS (any AFS will do, but I'm
> guessing OpenAFS is the most common.)

(for more in-detail questions you might want to use the openafs mailing 
list)

> Is it possible to have either / and / or:
> 
> Servers with replicas hosting a read/write volume.  One of the servers
> goes down, but the clients use the server that is still up...

read/write: currently not (there is some ongoing work but nothing of 
production quality yet).

read-only: certainly. However, this requires an explicit operation to 
sync the read-only replicas from the "master" read/write volume. In 
other words, this is not transparent to the users, but easily handled as 
part of any "release" procedure.

> Or one of the servers goes down, and the clients keep working on the
> local cached copy.

No (at least not for long). As soon as the client realizes that the 
server is gone, it will stop accessing data from it. The reason is that 
AFS caches need to be coherent across the system, and there are no 
provisions to synchronize state after a server outage.

"disconnected" operations were one of the key features of CODA, a 
successor to AFS. Unfortunately this project seems not quite ready for 
major deployments..

> I'm considering switching from NFS, but its a big step, and I'm trying
> to suss out what the advantages are...

Suggest to list your requirements...
NFSv4 might be an alternative to AFS nowadays, but old NFSes were 
unsuitable for wide-area access, and required you basically to trust all 
the clients.
  Regards
jan