基于NS2的改进队列管理算法及其实现

主动队列管理是网络拥塞控制的重要控制机制,其好坏关系到整体网络性能的优劣。传统的PID控制器结构简单,性能良好,但在过程中参数固定。由于神经元的固有优点,它能够有效克服PID控制器参数难以整定和面对非线性系统难以处理等缺陷。设计了单神经元自适应PID控制器。从2个方面进行了分析和验证:针对TCP拥塞避免的动力学模型,以控制理论为指导,在Matlab软件环境下,对传统的增量式PID算法和改进的单神经元自适应算法进行分析和仿真,利用队列长度和数据报丢失概率2个参数的性能优劣进行比较;脱离理论数学模型,采用NS软件中节点、链路和FTP业务源等网络元素搭建试验平台,将PI、PID和改进算法应用其中,通过观测队列长度变化和数据源链接数的改变,考察系统的稳定性和自适应性。仿真和实验充分说明改进的单神经元自适应PID控制器具有更好的稳定性、鲁棒性和自适应性。

Improved queue management algorithm based on NS2 and its implementation

Active queue management is an important congestion control mechanism,and its quality concerns the performance of whole network.The traditional PID controller has simple structure,good performance,but the parameters are fixed in whole process.The neuron can effectively overcome the disadvantages of PID controller in parameter adjustment and non-linear system.A single-neuron adaptive PID controller is presented,which is analyzed and tested in two aspects.Based on a developed dynamical model,which avoids TCP congestion,both the traditional PID controller and the improved controller are simulated in Matlab software environment and compared in queue length and data loss probability under the guidance of control theory.Without theoretic math model,the test platform is constructed using network elements of NS software,such as nodes,links,FTP service source,on which PI,PID and improved algorithms are applied to test system stability and adaptability by observing the changes in queue length and link number.Simulation results show the algorithm proposed better in stability,robustness and adaptability.