Heavy-traffic asymptotics for stationary GI/G/1-type Markov chains (Q439913): Difference between revisions

The paper studies heavy-traffic asymptotics for a stationary two-dimensional GI/G/1-type Markov chain describing a discrete-time single-server queue with general service time and Markov arrival process (phase-type process). The first component describes a level (queue-size) process while the second component describes the phase of the input process. The terminology is motivated by the embedded Markov chain method for classical non-Markov systems. It is proved, under some assumptions on the governing matrix of the input process, that diffusion-scaled level process converges weakly to a reflected Brownian motion as the mean drift in level goes to zero. These (irreducibility and moment) assumptions are weaker than that have been known earlier. To this general case, also, the following result is extended. The stationary distribution of the properly scaled level process is geometric and independent of the phase variable. Matrix analysis and the characteristic function are the main tools.