[OpenAFS] Re: read-only to read-write fast?

[Tino Schwarze]

On Mon, Aug 18, 2003 at 10:45:19AM +0200, Hein Roehrig wrote:

> >> Ideally, each workstation would become a AFS fileserver for a few user
> >> volumes. 
> >
> > Why do you want to do that? It makes your cell a lot more vulnerable
> > because the server keys need to be on each client. What are you trying
> > to accomplish by making clients a server?
> 
> Save $$$ for a fileserver... In our current setup there is no central
> big fileserver and users tend to have their home on "their"
> workstation so that the NFS automount most of the time just does a
> bind mount and gives local fs performance. The obvious downside of
> this approach is that (total) disk failures tend to destroy a day's
> worth of work... 

You won't get rid of this problem by using AFS. You'll still lose a
day's worth of work if the disk crashes. Remember: There's no live RW
replication, RO's need to be cloned manually (by calling "vos release").

> As was recommended somewhere, we would first migrate the ro stuff to
> AFS, like software on /opt. For this an AFS->NFS gateway should be
> straightforward, no? 

An RO-NFS-Translator is easy - just use a linux box with user-space NFS.

> For the home directories, we are going to put a few users on our
> (small) central server to gain experience. If this works out, then
> more hardware can be justified :-) Is anybody using rw AFS->NFS
> gateways? I read that the OpenAFS NFS-translator source should work on
> Solaris!? The point is that it will take some time to get an AFS
> client installed on all workstations...

I don't know about the AFS-NFS-Translator with authentication though...

Bye, Tino.

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