ABSTRACT

We illustrate by an example the potential of distributed
scheduling  with  {\em Active  Idleness}  to improve the
performance of  multi-class queuing  networks, which are
originally controlled by non-idling, or work-conserving,
policies.  The queuing network  we use in our simulation
experiments is due  to Dai, [1]. This particular network
was  shown  to be  unstable  if FIFO is  the  scheduling
policy.  However, for the  {\em Last Buffer First Serve}
policy, [2], it  is stable.  Under this setting  we show
that forcing  inactivity during  some periods of time in
the  presence  of  customers  may result  in significant
performance gains.