A two-level stochastic approximation for admission control andbandwidth allocation

摘要

In an access node to a multiservice network [e.g., a base stationnin an integrated services cellular wireless network or the optical linenterminal (OLT) in a broad-band passive optical network (PON)], thenoutput link bandwidth is adaptively assigned to different users andndynamically shared between isochronous (guaranteed bandwidth) andnasynchronous traffic types. The bandwidth allocation is effected by annadmission controller, whose goal is to minimize the refusal rate ofnconnection requests as well as the loss probability of cells queued in anfinite buffer. Optimal admission control strategies are approximated bynmeans of backpropagation feedforward neural networks, acting on thenembedded Markov chain of the connection dynamics; the neural networksnoperate in conjunction with a higher level bandwidth allocationncontroller which performs a stochastic optimization algorithm. The casenof unknown, slowly varying input rates is explicitly considered.nNumerical results are presented that evaluate the approximation and thenability to adapt to parameter variations