NSF Award
9660335
*** 9660335 Kwan This Small Business Innovation Research Phase I project describes a new approach to flow control in high speed communication networks which can significantly reduce the packet delays. In many networks, even the traffic is routed optimally, there are situations where the total traffic into the network is much larger than the network can handle. If no control strategies are present to limit the traffic into the network, queue sizes at bottleneck links will grow and packet delays will increase which may violate some maximum delay specifications for voice and video signals. The flow control problem is first treated as a fluid-flow dynamic system with time delay. Based on the model, we propose a flow controller that can generate some predictive information about the future queue length in the system. This predictive information is crucial to the success of the flow control scheme since it guarantees that the time delay of packets will not go unstable. Thus the maximum time delay performance criterion can be assured. In addition, the new flow control method can also maintain certain throughput of the system which is very important in applications such as video distribution. A new type of neural network called Fuzzy CMAC (Cerebellar Model Arithmetic Computer) is also proposed to predict the total information delay and the channel disturbance. The outputs of the Fuzzy CMAC will be passed to the flow controller to enhance the performance even further. The new flow control method can guarantee the maximum packet delay criterion and can also maintain a desired minimum throughput of the system which is very important in applications such as video distribution. Another advantage is that the controller is very easy to implement. The proposed algorithm should find many applications in broadband ISDN and ATM networks. ***
