Admission control and state dependent routing for multirate circuit-switched traffic

We investigate the general problem of admission control and alternate routing in a network for multirate circuit-switched traffic. One cannot assume that it is always beneficial to multiplex traffic classes that differ widely in their characteristics, without providing additional control mechanisms (such as bandwidth reservation for some classes). Similarly, the different traffic classes may require different flow-control and alternate routing rules in a network. For n traffic classes, the Markov decision process for even the single-link admission problem has an n-dimensional state-vector, and becomes numerically intractable even for small values of n. For the single-link admission problem, we propose an approximation that allows us to work with the scalar state variable consisting of the total bandwidth occupied by all the calls in progress on the link, regardless of the number of traffic classes. The simplified decision process lends itself to the one-step policy iteration procedure used by Krishnan and Ott for state-dependent routing of single-rate traffic. With our approximation, the multirate admission-control problem on a link is solved by a single system of linear equations of size (C+1), where C is the number of 'trunks' (integer multiples of a basic bandwidth unit) in the link. The call-admission criterion derived from this single-link analysis at once generalizes to a state-dependent rule for alternate routing and flow-control for multirate circuit-switched traffic in a network. We present simulation results comparing the performance of the proposed link-admission rule with that of the 'greedy' rule (which always admits every call that can be carried), and the performance of the proposed state-dependent routing and flow-control algorithm with that of a 'greedy' sequential routing algorithm. The results show blocking reductions in the range of 17-35% in the examples considered.