An Optimal RIO With Statistical Delay Assurances

Abstract

DiffServ uses the RIO active queue management technique to provide diferentiated service to the different aggre-gated traffic classes. The parameter setting of RIO dictates the behaviour of the router and thus the overall QoS performance of the network router. We develop an optimization framework that strives to guarantee statistical delays in a RIO router. We solve the optimization problem using nonlinear programming techniques, viz., penalty and barrier methods. The gradient and Hessian terms required in the optimization procedure are obtained using a novel adaptive four-timescale stochastic approx-imation algorithm and a modified second-order SPSA parameter update is then performed. We demonstrate the effectiveness of the framework by discussing the results of the simulations that we have performed. We observe that the proposed framework is an effective mechanism to ensure statistical delay guarantees while functioning with an optimal set of parameters in RIO routers.