考虑传统网络拥塞控制忽略了网络拥塞的持续状态,引入将数据包到达链路速率作为控制器输入的方案,得到一种改进单神经元梯度学习(improves single neuron gradient learning,ISNGL)的主动队列管理算法.ISNGL算法采用梯度学习动态调整网...考虑传统网络拥塞控制忽略了网络拥塞的持续状态,引入将数据包到达链路速率作为控制器输入的方案,得到一种改进单神经元梯度学习(improves single neuron gradient learning,ISNGL)的主动队列管理算法.ISNGL算法采用梯度学习动态调整网络参数,并在此基础上对收敛速度和稳定性加以改进,提出带有位移参数的新激活函数和带有权值调整的动量项的改进方法,最后通过NS2网络仿真软件在无线网络的拓扑模型上进行仿真分析,结果表明ISNGL算法在无线网络环境下拥有良好的拥塞控制能力.展开更多
Wireless mesh network （WMN） is a new multi-hop network for broadband accessing to intemet. However, there exists a server unfairness problem based on different hop distances in WMN. To solve this problem, the unfair...Wireless mesh network （WMN） is a new multi-hop network for broadband accessing to intemet. However, there exists a server unfairness problem based on different hop distances in WMN. To solve this problem, the unfairness issue was analyzed in test-bed experiment and NS2 simulation. A dynamic queue management scheme E-QMMN was proposed, which allocates the queue buffer according to the hop distance of every flow. The experimental results show that the proposed scheme can not only increase the hop distance fairness of the legacy scheme at most 50%, but also reduce the average round trip time at least 29% in congested WMN environments.展开更多
文摘考虑传统网络拥塞控制忽略了网络拥塞的持续状态,引入将数据包到达链路速率作为控制器输入的方案,得到一种改进单神经元梯度学习(improves single neuron gradient learning,ISNGL)的主动队列管理算法.ISNGL算法采用梯度学习动态调整网络参数,并在此基础上对收敛速度和稳定性加以改进,提出带有位移参数的新激活函数和带有权值调整的动量项的改进方法,最后通过NS2网络仿真软件在无线网络的拓扑模型上进行仿真分析,结果表明ISNGL算法在无线网络环境下拥有良好的拥塞控制能力.
基金Projects(61163060,61103204,60963022) supported by the National Natural Science Foundation of ChinaProject(D018023) supported by the Natural Science Foundation of Guangxi Province,ChinaPostdoctoral Funding of Central South University,China
文摘Wireless mesh network （WMN） is a new multi-hop network for broadband accessing to intemet. However, there exists a server unfairness problem based on different hop distances in WMN. To solve this problem, the unfairness issue was analyzed in test-bed experiment and NS2 simulation. A dynamic queue management scheme E-QMMN was proposed, which allocates the queue buffer according to the hop distance of every flow. The experimental results show that the proposed scheme can not only increase the hop distance fairness of the legacy scheme at most 50%, but also reduce the average round trip time at least 29% in congested WMN environments.
