一种共享存储交换结构中的双门限队列控制策略

A Dual Threshold Buffer Management Scheme in Shared Memory Switch Fabric

传统的共享存储交换结构的门限控制算法通常以当前各个队列长度为依据,缺乏对网络节点设备全局流量场景的考虑,且对组播的支持不足.本文提出使用有效业务量作为控制各个端口队列门限的主要依据,让各个端口承担相同的流量压力,从而使系统保持均衡状态.在经典的有效带宽理论的基础上,结合输入流量速率和分配的缓冲区大小一起来定义输出端口的有效业务量,给出了"缓存换带宽"的计算公式,对流量压力进行准确度量.双门限的使用使得共享存储空间既能够保持在平均意义下的平衡,又能实现突发时段的调剂.进一步,对组播信元的转发也可以纳入这个算法框架.模拟结果显示,在不同的流量模式下,这个算法比传统动态门限算法在取典型值α=1和组播浓度为30%时,对芯片的使用效率和端口的公平性分别提高15%和25%以上.

This paper concerns with the threshold control problem in a shared memory switch fabric. Traditional threshold control schemes are insensitive to the runtime traffic scenarios, and lack an efficient support for multicast forwarding. A novel scheme is presented in this paper, which aims to balance traffic pressure to the output ports and keep the system in an equilibrium state. Derived from a further exploitation of the developed effective bandwidth theory, the proposed scheme delineates an expressive formula that can calculate the shift of bandwidth burden to memory allocation. The association between bandwidth and memory utilization can be used to balance the traffic pressure by adjusting the per-queue thresholds in the shared memory.Furthermore,the presented framework can also encompass the support for multicast cells. The result of simulation shows that the scheme outperforms the traditional dynamic threshold in terms of efficiency and fairness under variant traffic modes.