A multiclass input-regulation throttle

Abstract

An input-regulation throttle to allocate capacity among multiple classes of jobs, where the allocation is enforced only when the total offered load is beyond capacity, is presented. The design uses a rate control throttle with a dedicated token bank for each class and a single, shared overflow bank. The per-class blocking and throughput are computed by alternative, approximate analyses. For the important special case of two priority classes of Poisson job arrivals, an exact analysis is given. The simple fluid approximation does surprisingly well, except in the case of roughly equal token and job arrival rates and small bank capacities. The stochastic Markov-chain-Poisson approximation yields only modest improvement over the fluid approximation. Thus, approximations that reasonably capture at least the second moment of the overflow processes are needed for accurate estimates of blocking for the important case in which the offered load is near the regulated limit. The two-parameter approximation with simulation for fine tuning seems adequate for most engineering purposes.