New Algorithms for Capacity Allocation and Scheduling of Multiplexed Variable Bit Rate Video Sources

This study presents simple and accurate heuristics for determining the equivalent bandwidth for multiplexed variable bit rate (VBR) video sources. The results are based on empirical studies of measurement data of various classes of VBR video sources. They are also validated through extensive simulation. The principal result is that the equivalent bandwidth per source for n independent and identically distributed VBR video sources may be approximated by a hyperbolic function of the form: a coth -1n + b where a and b are independent of n. Further, assuming ∈ is the acceptable loss tolerance, statistical regression shows that b is a linear function of mean and log ( ∈ ), while a is a polynomial in log( ∈ ). The capacity assignment problem is further augmented with a scheduling algorithm that is an extension of the Virtual Clock Algorithm. The new algorithm belongs to a class of algorithms which we refer to as Generalized Virtual Clock (GVC) algorithms. The particular GVC algorithm investigated in this paper estimates the instantaneous rate of transmission of each source, and uses the estimate instead of the static average rates, for prioritizing packets. In so doing, it attempts to synchronize the switch scheduling rates and the packet arrival rates of each source, and improves upon the spatial loss distribution characteristics of Virtual Clock. The combined allocation and scheduling algorithms are proposed as means for guaranteeing Quality of Service in high speed networks