Fluid model for access control mechanism

Abstract

In this thesis, we develop two distinct traffic models: one based on Brownian motion and the other based on fractional Brownian motion. The later model captures the self-similarity and the long-range dependence (LRD) properties. The aggregate model is composed of a drift part and of a fluctuation (diffusion) part. With this model, traffic from several seconds to 24 hours can be simulated. Applying the Token Bucket (TB) mechanism, a continuous time (state-space) dynamic system model is developed based on the ideas of recent papers [1,2]. Incoming traffic from each user is policed at the TBs and one multiplexor buffer, linked to all the TBs, multiplexes the conforming traffic. We propose two feedback control strategies, one is a simple feedback control law and the other is a feedback control based on neural network, to control the traffic flow into the link of the backbone network. We also use the simulated annealing algorithm to optimize the parameters of control laws. Several network performance related issues are studied systemically. The results show that the proposed control laws can improve the network performance, by improving throughput, reducing multiplexor and TB losses, and relaxing, not avoiding, congestion.