* -DDS Daemon Synchronization

      Normally, each ntserv process keeps its own timer and updates its client
      when SIGALRM is received.  Each time the player dies, the timer is reset.
      This means that there's no way to predict when a ntserv process will
      update its client.  It could be at the same time the daemon is running
      move() or just before, effectively giving the client information that
      is almost immediately out of date, or it could fall in the interval after
      the daemon completes move().  Likewise, there's no way to predict how
      many ntserv processes are active simultaneously.

      With daemon synchronization, the daemon signals the ntserv process
      when it completes move().  This helps guarantee that the ntserv
      processes are updating the client only after the daemon is finished
      with that cycle and ensures that the clients get the most recent
      changes immediately.  Since the ntserv processes usually run at
      5 update per second, the daemon need only signal half the processes
      on each update.

      DS adds two new fields to the player structure:

         int    p_process;
         int    p_timerdelay;

      The daemon signals a ntserv process by sending a SIGALRM to that
      process id.  The process number also functions as a way to kill
      a ntserv process if the daemon dies or if a player status goes to
      PFREE (note: if status goes to PFREE and the ntserv process is not
      killed, another person can join that slot creating the co-pilot problem).
      The FREE command also kills the ntserv process, guaranteeing that the
      slot is free and will remain free.

      The only known drawback of this approach is that if the daemon slows
      down greatly or stops (with the exception of PAUSE), client updates
      decrease or stop.  This is normally only noticeable when the daemon
      runs 'end_tourney'.

      The '*' client command can be used on an individual basis to toggle
      between daemon synchronization and using a process-local timer
      (original method).

      The gain of the approach is in eliminating most of the competition
      between the daemon and the ntserv processes for the cpu and
      shared-memory during the crucial daemon update cycle, ensuring that
      all clients get the most up-to-date information possible for their
      update time, and evening out server machine performance by interleaving
      processes.

      This method is being used on calvin.usc.edu (ack, bad example) and
      vlsi.ics.uci.edu.


-From Tedd Hadley's note in INL server