This paper proposes a framework for optimal policing in ATM networks.
A general approach to modeling controllers which implement the policin
g function is described, using finite-state automata. The theory of fi
nite-state accepting automata and regular expressions enables the expl
icit representation of all traffic patterns which pass through the pol
icing controller, without modification. This in turn enables the symbo
lic computation of the worst-case source and therefore provides a unif
ied framework for the formulation and derivation of optimal bandwidth
enforcement schemes. Exhaustive searches are performed for small cases
to find optimal finite-state automata. These optimal results are exte
nded to obtain near-optimal solutions for larger-state automata. The o
ptimal and near-optimal policing controller designs are compared to co
nventional schemes, such as leaky buckets, and improvement in performa
nce is achieved, where performance is measured in terms of cell loss r
ate for the declared source and throughput of the corresponding worst
case source.