Parijat Dube



Integrated Services Packet Networks support a guaranteed transport service (also called stream traffic) that is used by applications that have an intrinsic (unmodifiable) temporal behavior, and hence have tight end-to-end delay requirements (e.g. real-time Variable Bit Rate (VBR) services).

Most on-line (i.e., not stored) VBR sources (like packet telephony and motion compensated variable rate coded video) would find it difficult to a priori declare the traffic parameters required for a connection admission control strategy. There is thus the problem of measurement based on-line estimation of source parameters. In this talk we address the problem of on-line estimation of OPTIMAL source parameters. The parameters are OPTIMAL in the sense that they simultaneously:

1. Minimise a network resource(buffer at the switches + link bandwidth) cost function. 2. Bound the shaping plus the multiplexing delay.

We provide a formulation for the characterisation of optimal leaky bucket shaper parameters (since a source is invariably shaped before feeding it into the network), i.e., the optimal leaky bucket rate parameter (rho*) and the optimal leaky bucket depth (sigma*). We then develop measurement based on-line algorithms for estimating the opimal parameters, rho* and sigma*. The algorithms are based on stochastic approximation (continuous as well as discrete) techniques and can be implemented at the source and/or the switch for on-line estimation of OPTIMAL shaper parameters. Our approach yields significant reduction in network cost by working with these on-line estimated OPTIMAL parameters. The research was supported by NORTEL NETWORKS, USA.

[Parijat Dube]