[MPICH] MPD instructions for root

[Narayan Desai]

We've run mpd in a single root ring mode on one of our clusters for a
few years. Basically, there are a few caveats. The ring recovery
algorithm counts on being able to reconnect to its initial connection
point upon failure events. In a user ring, this location is transient,
and is handled by mpdboot. In a persistent ring, we typically pick a
single location and fixed port (usually on the head node) where a
master mpd daemon is started. All other nodes are run with either
config or command line options to connect into the mpd ring at that
fixed location. 

The only other gotcha is that you need to ensure that mpdroot has the
proper permissions; it assumes you are running a root ring if mpdroot
is setuid root, iirc. Other than that, it should run just like a user
ring. 
 -nld