> Hey Stephen,
> 
> > Hi,
> > 
> > On Wed, 14 Dec 2005, John A. Goebel wrote:
> > 
> > >++ 14/12/05 18:23 +0000 - <John Rowe>:
> > >
> > >Hi,
> > >
> > >>>What do you want to do with this filesystem?
> > >>
> > >>Aways a good question. I have a bunch of nonidentical  PCs running SL4.x
> > >>connected by gigabit. People can log into any of them and see their home
> > >>space. Currently I do this by having two of them as file servers, each
> > >>with a mirrored disk pair. Ideally I would like to have a single virtual
> > >>"/home" filesystem which I could add physical disks to and which is
> > >>resilient to any one node being down.
> > >
> > >I don't know what your load is like, but can't NFS do this for you and LVM?
> > >
> > >For the requirement of serving $HOME, NFS is a classic. Although I haven't
> > >tried it, NFS 4 has failover. Maybe someone else has a better suggestion?
> > 
> > Even if that's implemented now (is it?), wouldn't you still need some 
> > (cluster) filesystem shared between the servers?
> 
> Yes, that's right. Years ago, when I had to do this, I used 
> Kimberlite. Sorry, I don't know the current resources.
> 
> I like the simple solution if it works, so here a recipe:
> 
> http://www.linux-ha.org/HaNFS
> 
> > AFS won't do the job either (no read-write replication). It would be 
> > possible to recover from a failed node quickly and transparent to the 
> > client, though.
> > 
> > The most promising solution in such a scenario to me seems to be something 
> > like that described in
> > 
> >   http://www.rhic.bnl.gov/hepix/talks/041020am/miers.pdf
> > 
> > Not trivial to set up, and I wonder how lock recovery would work, but the 
> > best low cost solution I know of.
> 
> Also, you might want to poke around Linux HA site. The LAN mirror section
> especially:
> 
> http://www.linux-ha.org/RelatedTechnologies_2fLanMirroring
> 
> John

Just to add my 2 cents worth, I use linuxha.net (http://www.linuxha.net) which
uses drbd as the shared storage device, so you don't have to spend the $k's on
shared storage hardware, but use a network link between 2 nodes to replicate
data. This also very easily provides HA NFS, using failover between two machines.

Michael.