Active queue management design using discrete-event control

Recently, control-theoretic approaches have been studied and employed to manage and control the performance of computing systems. Most of the existing control-theoretic approaches model computing systems as linear systems and apply feedback control. In this paper, we show discrete-event modeling and control techniques can be effectively applied to performance management and control of computing systems. We use active queue management design for Internet routers as a running example. By modeling the router as a queueing system, we formulate the problem of designing the optimal dropping strategy as an optimal queueing control problem under discrete-event control framework. We then derive the optimal controller design using uniformization and value iteration. Through numerical evaluations, we also discuss the effect of various design parameters and workload characteristics on the optimal dropping strategies derived.