Algorithms and architectures for packet classification

根据行业领导者的销售数据估计，全球路由器和交换机的销售额每年在200亿至300亿美元之间。该计划的目的是推动最先进的数据包分类，这是在这些设备中使用的基本技术。数据包分类通常是将数据包中选定字段的内容与一组过滤条件进行比较，决定数据包是否匹配这些条件中的一个或多个，然后在一个或多个匹配的情况下选择最佳操作。近期的前景是每秒运行兆比特的链路，以及允许数亿兆比特的新的寻址形式。这意味着分类器将不得不处理每个链路每秒数十亿个数据包，过滤器集将从数万条规则增长到数百万条规则。为了实现这一目标，需要新的数据包分类算法和实现。

Global sales of routers and switches, estimated by looking at the sales figures of the industry leaders, are in the range of $20 billion to $30 billion annually. The objective of the proposed program is to push the state of the art in packet classification, which is the fundamental technique used in these devices. Packet classification, very generally, is the operation of comparing the contents of selected fields in a packet to a set of filter conditions, deciding whether the packet matches one or more of those conditions, and then selecting the best action in the case of one or more matches. The near term prospects are for links running at terabits per second, and new forms of addressing that will allow for hundreds of millions of subnets. This means that classifiers will have to deal with billions of packets per second per link, and that filter sets will grow from tens of thousands of rules to millions. New algorithms and implementations of packet classification are required to enable this future.