Off-the-path flow handling mechanism forhigh-speed and programmable traffic management
Off-the-path flow handling mechanism forhigh-speed and programmable traffic management
复制标题
用于高速和可编程流量管理的路径外流量处理机制
DOI：
--
复制
发表时间：
2008
期刊：
影响因子：
--
通讯作者：
A. Iwata
中科院分区：
文献类型：
--
作者：
H. Shimonishi;T. Yoshikawa;A. Iwata
在本文中，我们提出了一个高速和可编程的流量管理机制，使容易和及时的创新。由4D、Tesseract或OpenFlow引入的控制框架将控制功能从交换机节点分离到控制服务器，使得可以在交换机外部实现各种网络控制策略。在此框架内，我们提出了一种机制，使灵活的基于流量的交通管理，使各种创新的交通管理计划可以实现。然而，每个流的流量管理需要逐包的状态更新，这可能破坏这种控制框架。所提出的机制由控制服务器和交换机组成，控制服务器使用从交换机发送的采样数据包来监控流量状况并计算每个流的数据包丢弃率，交换机根据丢弃率丢弃传入的数据包。分组采样和丢弃不需要在交换机处逐个分组的状态处理，并且因此允许来自控制服务器的控制。我们还提出了一种机制，使用时间序列的布隆过滤器来压缩丢弃信息，以便允许频繁的控制更新。我们测试的机制与每流WFQ仿真和仿真结果表明，很好的每流公平性。此外，我们发现流表被压缩了600倍，并且服务器和交换机的处理成本足够小，可以用于10 Gbps链路。
In this paper, we propose a high-speed and programmable traffic management mechanism to enable easy and timely innovations. A control framework introduced by 4D, Tesseract, or OpenFlow, separates control functions from the switch nodes to a control server so that a variety of network control policies can be implemented outside of the switches. Within this framework, we propose a mechanism to enable flexible flow-based traffic management so that a variety of innovative traffic management schemes can be realized. Per-flow traffic management, however, requires packet-by-packet state updates, which can spoil this control framework. The proposed mechanism consists of a control server that monitors traffic conditions using sampled packets sent from the switches and calculates per-flow packet discarding rate, and switches that discard incoming packets according to the discarding rate. Packet sampling and discarding do not require packet-by-packet state handling at the switches and thus allows controls from a control server. We also propose a mechanism to compress the discarding information using a time series of bloom filters, so that frequent control updates are allowed. We tested the mechanism with per-flow WFQ emulation and the simulation results showed very good per-flow fairness. Furthermore, we found that the flow table is compressed 600 times smaller and that the processing cost at the server and the switches is small enough for use with 10 Gbps links.