万兆网络中拥塞控制和排队调度实现与优化

Implementation or optimization of congestion control and queue scheduling allocation based on 10G networks

研究了万兆交换网络中交换原理,对万兆交换机中性能的关键因数和瓶颈存在于缓存区中的拥塞控制和排队调度机制,基于早期的网络应用传统的拥塞控制及避免,排队调度方法能很好的处理网络时延和丢包问题。然而对于万兆网络无法达到很好的性能要求,因此有必要寻求新的拥塞控制及避免和排队调度算法。本文在32口万兆交换网络中应用了端到端快速重传快速恢复(E2E-CC)拥塞控制机制,加权的早期随机检测(WRED)拥塞避免机制,多优先级加权差额循环调度(WDRR)机制,最后交换机正常工作时实现了交换大包数据(1 518 bytes)时延小于3μs,小包数据(64 bytes)时延小于2μs,丢包率在10亿分之一以下的性能,并且提供端口流镜像、流统计、VLAN(虚拟网络)划分、万兆链路聚合,抑制广播风暴等服务质量(QoS)功能。

After studying the theory and implement based on 10 G switch networks,there are a key factor and a bottleneck in the congestion control and queue scheduling mechanism. The early solution can solve the latency and loss rate problems. But the early mechanism and scheduler applying in the 10 G switch networks can't work very well. congestion control and resource allocation new mechanism have been applied in this paper. This including End-to-End Congestion Control mechanism(E2E-CC),Weighted Random Early Detection mechanism(WRED) and more Priority Weighted Deficit Round Robin mechanism(WDRR). Finally,the operation of the switch networks with 32 ports have less 3 μs latency on big packet(1 518 bytes),or less 2 μs latency on small packet(64 bytes),less one billion points packet loss rate and with Mirrored Traffic/Traffic Metering/VLAN Shaping quality of service function and so on.