RIA: Trading Packet Headers for Packet Processing

华盛顿大学圣路易斯分校乔治瓦格斯RIA：交换数据包报头进行数据包处理 在高速网络中，分组处理相对昂贵，而带宽便宜。 这里的中心论点是，如果合适的报头字段可以添加到数据包中，那么可以设计出加快数据包处理的机制。 本文提出了若干观点来支持上述论点。 首先，数据操作（例如数据复制、校验和）是终端系统分组处理的主要瓶颈。 这里采用的方法建议将数据操作报头添加到每个数据包的易于访问的部分。 这个头包含指向数据操作所需的字段（在各个层中）的指针。 该信息（例如，指向要加密的数据的指针）允许实现有效地联合收割机组合数据操纵步骤（例如，加密和复制）。 先前的工作已经表明，组合数据操作步骤可以产生数量级的性能改进。 本方法可以以更统一和结构化的方式产生类似的改进。 其次，工作研究在所有层（例如连接标识符，网络地址）的协议标识符中添加索引字段，以减少查找成本和通用协议处理。 研究了利用这些索引字段的几种新方法（线程索引、索引传递和源哈希）。 已知在虚拟电路分组上使用虚拟电路标识符（VCI）简化了查找和分组处理。 在源散列和线程索引中，添加的索引本质上用作VCI，但用于数据报网络中的流。 例如，在源散列中，“VCI”是由源选择的一致的随机标签。 我们的新方法提供了普通VC的优点，而不需要建立的往返延迟。 这些方法可以将最优情况下的数据报查找时间从O（log（n））降低到O（1），这对于千兆路由器来说是很重要的。 在这个项目中，我们设计，分析和实现这些新的想法，并寻找其他技术，从基础论文中出现。 当前的过渡环境（其中传输、路由和数据链路协议正在发生变化）提供了应用这些技术来影响真实的协议的机会。

Washington University at St. Louis George Varghese RIA: Trading Packets Headers for Packet Processing In high speed networks, packet processing is relatively expensive while bandwidth is cheap. The central thesis here is that if suitable header fields can be added to packets, then mechanisms to speed up packet processing can be devised. Several ideas are advanced to support the above thesis. First, data manipulation (e.g. data copying, checksumming) is a major bottleneck of end system packet processing. The approach taken here suggests adding a data manipulation header to an easily accessible portion of each packet. This header contains pointers to fields (in various layers) required for data manipulation. This information (e.g. a pointer to data to be encrypted) allows implementations to efficiently combine data manipulation steps (e.g. encryption and copying). Prior work has shown that combining data manipulation steps can yield order-of-magnitude performance improvement. The present approach can yield similar improvements in a more uniform and structured fashion. Second the work studys adding index fields to protocol identifiers at all layers (e.g. connection identifiers, network addresses) to reduce lookup costs and generic protocol processing. Several new ides to utilize these index fields (threaded indexing, index passing, and source hashing) are studied. It is known that the use of Virtual Circuit Identifiers (VCIs) on virtual circuit packets simplifies lookup and packet processing. In source hashing and threaded indexing, the added indices essentially serve as VCIs, but for flows in a datagram network. In source hashing, for example, the "VCI" is a consistent random label chosen by the source. OUr new methods provide the benefits of normal VCis without requiring a round trip delay for set up. The methods can lower uorst case datagram lookup times form O(log(n) to O(1), which may be important for gigabit routers. IN this project we design, analyze, and implement these new ideas and search for other techniques that arise from the basis thesis. The current climate of transition (in which transport, routing, and data link protocols are changing) provides an opportunity to apply these techniques to influence real protocols.