ECONOMIES OF SCALE FOR LONG-RANGE DEPENDENT TRAFFIC IN SHORT BUFFERS

Consider the problem of predicting loss ratios for traffic streams sha ring a buffer. Approximations based on the temporal statistical proper ties of single sources do not account for the economies of scale which can arise when there is statistical multiplexing gain across sources. These can occur whether the sources have long or short range dependen ce; in either case the economies may be positive or negative. In this paper we investigate this matter for sources described by fractional A RIMA processes, and show that their short-range structure can mean tha t a simple power-law tail based on the Hurst parameter alone can be ov er-optimistic when the buffer space allocated per source is not large.