On stationary tandem queueing networks with job feedback

Nicholas Bambos, Kimberly M. Wasserman

Abstract: The class of tandem queueing networks with job feedback is studied under stationarity conditions on the arrival and service times sequences. Each job, after completing service in the last queue, is fed back (rerouted) to the first one, a random number of times, before leaving the system. The average execution time per job is exactly computed, as the number of jobs becomes large, and is minimized under mild conditions. The degree of parallelism achieved in the processing is also computed. The issue of rate-stability of the system is then considered. The network is defined to be rate-stable iff the job departure rate is equal to the job arrival rate; that depends heavily on the dynamic feedback policy we employ to place rerouted jobs in specific places of the front queue buffer of the network. The condition under which the network is rate-stable is specified, and a dynamic feedback policy is constructed, which rate-stabilizes the system under the maximum possible job arrival rate; thus, it maximizes the dynamic throughput of the network. Other related results concerning the performance of tandem networks with feedback are obtained.