> The problem with this is that you can have 2 machines start at the
> same time when the network is down. They both use the same address
> space and stuff everything up.
Not really a problem if you choose to see it like a process. If you force
a process to dump core, you can restart it later with gdb, also on another
machine. The same mecanism can be used to move processes between servers.
The actual decision and move mecanism don't even have to be in kernel
space, it can be implemented in user space.
> To do this properly you would need to have /dev/* refer to devices on
> the machine the process started on, and have /localdev/* have the files
> that are usually in /dev, and have /remotedev/machinename/* refer to
> remote devices. /dev/* could be some procfs like file system to support
> this.
Why not have a sort of common /dev dir in the cluster? When a new server
joins the cluster it get its dev files added to the common /dev (renamed
accordingly of course). Then no one will get confused when moved to a new
server.
Baldur