| Along with the Internet scale's unceasing expansion, on Internet's user quantity's swift growth, the network has had the congestion, which will lead to the congestion collapse. Once the network congests, the network performance will reduce seriously. The network throughput and efficiency become lower, the router buffer queue and packet time delay increase seriously, even the packets will be lost. And the jitter of time delay become intensely.According to the position from the control mechanism, the control algorithms may divide into two broad headings: Based on source terminal control algorithms and based on correspondence subnet control algorithms. In the source terminal control algorithms, the TCP agreement congestion control algorithm is most widespread, which is also called the TCP congestion control; the communication subnet's control policy (for example router and switchboard) carries out in the network equipment, which is also called the IP congestion control. And the active queue management is based on the correspondence subnet IP congestion control.This article mainly focuses on the active queue management algorithm, The benefits of AQM algorithms is that they can reduce the number of packets discarded in the router, make time delay of Interactive services a little lower, and avoid the global synchronization .At first, this article introduces the RED,ARED,REM, and PI algorithm. The RED algorithm can not estimate the seriousness of congestion. When there are lots of active TCP flows, because of sudden flows high, this algorithm will drop lots of packets, lead to short of network resource. The PI and PID algorithms can regulate the instantaneous queue lengths in router buffer more effectively, improve network resource utilization, but parameters setting depend on several times of experiment, so system transient and steady-state performance can not be a reliable guarantee. What's more, because the parameters are fixed, they can only adapt to limited network environment.This article has a deeply research on the PI algorithm and PID algorithm. The parameters of both the algorithms are set separately by using theory of optimal second-order model and theory of stability margin. And the performance of those algorithms is analyzed by simulation. At last, combining the advantage of PID algorithm and fuzzy control, P-Fuzzy PID algorithm is designed. The main research work includes the following points:A. Parameter settings of PI algorithm and performance analysisIn 2001, C.Hollot et use the classical control theory to design the AQM algorithm. They designed PI algorithm by linear treatment of TCP flow control model with Small-signal theory. In this article, base on the research of PI algorithm, we convert the PI controller to discrete form of controller by bilinear transformation. Then we set the parameters of the PI algorithm by using theory of optimal second-order model. In the Network Simulator of NS-2, the parameters of PI algorithm are gotten by many times experiment. So it lacks of theoretical basis, and the regulation time is very long. But the new parameters can make the instantaneous queue lengths in the router buffer close to the Goal queue length more quickly. The packet's drop ratio is lower and link utilization is higher.B. Parameter settings of PID algorithm and performance analysisAccording to Simulation results of PI algorithm, we can see it can work stably and effectively, but we also find out that the PI controller very depends on the router's buffer, and when time delay become longer, it's performance become bad. The direct cause might be the controller lacks of a differential link. So we search on the PID algorithm, get the digital PID controller by bilinear transformation, and set the parameters by theory of stability margin. It has a better performance in making the instantaneous queue lengths in the router buffer close to the Goal queue length. But in large load environment, the queue is hard to close to the goal length and unstable. Because the parameters are fixed, it can not adapt to the variable network environment.C. The design of P-Fuzzy PID algorithm and performance analysisThe main purpose of P-Fuzzy PID algorithm is to combine the advantages of PID algorithm and fuzzy control. The design idea is that when error is big, it choose P controller, the set of parameters mainly concern about response speed, to reduce error quickly. When error reaches the specified threshold, this controller switches to the Fuzzy PID controller, in order to realize express and stability of transient process, finally get excellent stability.The P-Fuzzy PID algorithm has good adaptive capacity and control effect to non-linear complex systems and large load systems. In large load environment, it can still make sure of that the instantaneous queue length in the router buffer close to the Goal queue length quickly. And it has higher link utilization. In environment with variable load and time delay, it also has good adaptability. But the set of parameters is complex which needs to be further optimized.D. ConclusionsThe PI and PID algorithms have good control effects in little time delay and light load network environment, and the queue stability is better. But the parameters is hard to fix, and it can not adapt to variable network environment, so we made P-Fuzzy PID algorithm, shows better control in larger time delay network environment that it can make the transient queue close to the goal queue more quickly, and there is no big turbulence. The drop ratio is also lower.This article simulates those three algorithms by NS2 network. And it proves the performance of P-Fuzzy PID algorithm is better than PID algorithm. |
PDF Full Text Request