Optimum feedback strategy for access control mechanism modelled as stochastic differential equation in computer network (Q2566898)

The paper considers a control strategy for computer communication network. A continuous-time stochastic traffic (source) model governed by a stochastic differential equation is developed, which captures main network traffic characteristics. The model is combined with the system describing the dynamics of the token bucket (TB) access control mechanism and the multiplexor. An objective function reflecting the losses at the TBs and the multiplexor is used to evaluate the performance of the system. A simplified feedback control law based on neural network is proposed and simulated annealing (SA) vs. recursive random search (RRS) is applied to optimize the control law (network) parameters. It is demonstrated by means of simulation experiment that RRS algorithm can save significant computation time without compromising the optimality. The numerical results show that the proposed control law can improve the network performance compared with standard open-loop strategies used in current network management practice.