Policing and traffic shaping at the user-network-interface (UNI)

Abstract

A network's decision to accommodate a new ATM connection is based in part on sustained and peak cell rate and peak duration declared by the user in the connection request. As these traffic descriptors are onlyprojections of the traffic, connection admission control must be supplemented by input rate control in order to monitor and regulate theactual traffic. None of the proposed input rate control schemes enforces multiple traffic parameters. Additionally, traffic shaping at the UNI based on regulating average and peak rates may enhance statistical multiplexing performance. This paper presents a performance analysis of the two-stage entry monitor combining a sticky bufferpolicer with aspacer. To the best of our knowledge,no analysis of tandem policers has appeared before this paper. The first stage enforces the sustained rateR/T negotiated in the connection request; the spacer shapes the traffic and controls burstiness by further ensuring that cells enter the network at leastS slots apart. We derive the probability generating function for the combined queue length distribution and evaluate numerically (i) the manner in whichS and burstiness affect the cell-loss probability, and (ii) the performance penalty caused by enforcing two rates. The principal contribution of this paper is the calculation of the buffer size required to police two parameters subject to a specified cell-loss ratio. Our principal conclusion is that the increase in buffer size needed to police the peak sustained rate in addition to the sustained rate is minimal. We will also demonstrate that the leaky bucket and the single-stage sticky buffer have comparable performance, enabling us to conclude that a two-stage sticky buffer can police both peak and sustained rates for almost the same cost as a leaky bucket.