[OpenAFS-devel] openafs cell and contrib area?

[Default]

> Well, one thing that might be nice to do here would be the idea of a
> 'pull replicate' instead of a push replicate. Somehow allow anyone with
> the appropriate permissions on a volume to PULL the volume from a remote
> server. The remote server needn't even be aware that the replicate
> exists, other than the initial validation of permissions. 

For the sake of the failover code, it helps that there are only ever two 
versions of a volume in existence at any one point in time, and that state 
really shouldn't exist for very long.  Redundancy suffers quite a bit when 
there are two versions extant, as failover must only be permitted going 
forward in time, never backward.

Push ensures that the oldest version goes away ASAP.

> Actually, something I've wished afs had for a long time is the concept
> of a server side cache. For example - I've got a bunch of local afs
> servers with big disks on them that aren't used. It would be nice if the
> afs clients could somehow ask the LOCAL file servers to get a callback
> to a particular file on their behalf, and then that client could then
> retrieve it from the cache on the server instead. i.e. allow AFS servers
> to proxy AFS servers. 

Sounds like IFS.  Or XFS.  Dig around and see if you can find the papers.


> The benefit would be that if you have 10 machines in the cell accessing
> transarc, you don't have to sit and wait for transarc's network/server
> on all 10 of them, just the first one to access a particular file.