网络切片可编程数据平面模型
2024,46(5):200-208
刘忠沛
国防科技大学 计算机学院, 湖南 长沙 410073，747541120@qq.com,gflv@163.com
吕高锋
国防科技大学 计算机学院, 湖南 长沙 410073，747541120@qq.com,gflv@163.com
王继昌
国防科技大学 计算机学院, 湖南 长沙 410073
杨翔瑞
国防科技大学 计算机学院, 湖南 长沙 410073
国防科技大学 计算机学院, 湖南 长沙 410073，747541120@qq.com,gflv@163.com
吕高锋
国防科技大学 计算机学院, 湖南 长沙 410073，747541120@qq.com,gflv@163.com
王继昌
国防科技大学 计算机学院, 湖南 长沙 410073
杨翔瑞
国防科技大学 计算机学院, 湖南 长沙 410073
摘要：
可重构匹配表(reconfigurable match table, RMT)是一种可编程的数据包处理流水线架构。为了实现可编程数据平面对更多不同网络协议的支持,在该架构的基础上扩展逆解析器,使用扩展后的逆解析器以及两个RMT流水线组成一个协议无关的网络切片可编程数据平面模型。由于RMT架构中采用精简指令集,扩展后的逆解析器采用复杂指令集,因此称扩展后的架构为混合指令RMT(hybrid-instruction RMT, HiRMT)。HiRMT能够支持基于IPv6转发平面的段路由、SID(segment ID)的多语义、微分段技术、多协议标签交换技术,以及虚拟扩展局域网技术。该架构具有广阔的应用场景。在Corundum原型平台上进行了逆解析器模块的性能测试,结果表明扩展后的逆解析器能够使用较少的资源在数据包大小达到512 B时以100 Gbit/s的吞吐量进行处理。
可重构匹配表(reconfigurable match table, RMT)是一种可编程的数据包处理流水线架构。为了实现可编程数据平面对更多不同网络协议的支持,在该架构的基础上扩展逆解析器,使用扩展后的逆解析器以及两个RMT流水线组成一个协议无关的网络切片可编程数据平面模型。由于RMT架构中采用精简指令集,扩展后的逆解析器采用复杂指令集,因此称扩展后的架构为混合指令RMT(hybrid-instruction RMT, HiRMT)。HiRMT能够支持基于IPv6转发平面的段路由、SID(segment ID)的多语义、微分段技术、多协议标签交换技术,以及虚拟扩展局域网技术。该架构具有广阔的应用场景。在Corundum原型平台上进行了逆解析器模块的性能测试,结果表明扩展后的逆解析器能够使用较少的资源在数据包大小达到512 B时以100 Gbit/s的吞吐量进行处理。
基金项目：
国家重点研发计划资助项目(2020YFB1805603)
国家重点研发计划资助项目(2020YFB1805603)
Network slice programmable data plane model
LIU Zhongpei
College of Computer Science and Technology, National University of Defense Technology, Changsha 410073, China,747541120@qq.com,gflv@163.com
LYU Gaofeng
College of Computer Science and Technology, National University of Defense Technology, Changsha 410073, China,747541120@qq.com,gflv@163.com
WANG Jichang
College of Computer Science and Technology, National University of Defense Technology, Changsha 410073, China
YANG Xiangrui
College of Computer Science and Technology, National University of Defense Technology, Changsha 410073, China
College of Computer Science and Technology, National University of Defense Technology, Changsha 410073, China,747541120@qq.com,gflv@163.com
LYU Gaofeng
College of Computer Science and Technology, National University of Defense Technology, Changsha 410073, China,747541120@qq.com,gflv@163.com
WANG Jichang
College of Computer Science and Technology, National University of Defense Technology, Changsha 410073, China
YANG Xiangrui
College of Computer Science and Technology, National University of Defense Technology, Changsha 410073, China
Abstract：
RMT (reconfigurable match table) is a programmable pipeline architecture for packet processing. In order to enable the programmable data plane to support more different network protocols, the deparser based on RMT was extended. A protocol-independent network slicing programmable data plane model was formed by using the extended deparser and two RMT pipelines. Since reduced instruction set was used in RMT architecture and complex instruction set was used in the extended deparser, the extended architecture was called HiRMT(hybrid-instruction RMT). HiRMT can support segment routing IPv6, multiple semantics for SID(segment ID), micro SID, multi-protocol label switching and virtual extensible local area network. This architecture has broad application prospects. The performance of the deparser module was tested on the Corundum prototype platform, and the results show that the extended deparser can process the packet size up to 512 B with a throughput of 100 Gbit/s with fewer resources.
RMT (reconfigurable match table) is a programmable pipeline architecture for packet processing. In order to enable the programmable data plane to support more different network protocols, the deparser based on RMT was extended. A protocol-independent network slicing programmable data plane model was formed by using the extended deparser and two RMT pipelines. Since reduced instruction set was used in RMT architecture and complex instruction set was used in the extended deparser, the extended architecture was called HiRMT(hybrid-instruction RMT). HiRMT can support segment routing IPv6, multiple semantics for SID(segment ID), micro SID, multi-protocol label switching and virtual extensible local area network. This architecture has broad application prospects. The performance of the deparser module was tested on the Corundum prototype platform, and the results show that the extended deparser can process the packet size up to 512 B with a throughput of 100 Gbit/s with fewer resources.
收稿日期：
2022-05-26
2022-05-26
