On Modeling Datagram Congestion Control Protocol and Random Early Detection using Fluid-Flow Approximation

摘要

In this paper, we model DCCP congestion control mechanism and RED as independent discrete-time systems using fluid-flow approximation. By interconnecting DCCP connections and RED routers, we model the entire network as a feedback system called DCCP/RED. We then analyze the steady state performance and the transient state performance of DCCP/RED. Specifically, we derive the packet transmission rate of DCCP connections, the packet transmission rate, the packet loss probability, and the average queue length of the RED router in steady state. Moreover, we investigate the parameter region where DCCP/RED operates stably by linearizing DCCP/RED around its equilibrium point. We also evaluate the transient state performance of DCCP/RED in terms of ramp-up time, overshoot, and settling time. Consequently, we show that the stability and the transient state performance of DCCP/RED degrade when the weight of the exponential weighted moving average, which is one of RED control parameters, is small. To solve this problem, by adding changes to the function with which RED determines the packet loss probability, we propose RED-IQI (RED with Immediate Queue Information), as an applications of our analytic result. We analyze the transient state performance of the feedback system DCCP/RED-IQI where DCCP connections and RED-IQI routers are interconnected. Consequently, we show that DCCP/RED-IQI has significantly better transient state performance than DCCP/RED.