A Study Of GPU Accelerated Packet Processing

With the rapid development of Internet, there are higher demands on routers. Firstly, routers shall provide sufficient packet forwarding performance. Secondly, router shall be extensible so that ISP can provide new services based on them. Traditionally, the extensibility of the network infrastructure is achieved through additional hardware or middleboxes like NAT box. Although state-of-the-art routers are pursuing their exten-sibility through increasing their programmability, the nature of the specialized hard-ware and software prevents hardware-based routers from being the most extensible solution.Building routers with commodity software and hardware is an attractive alterna-tive approach. This approach has great programmability, extensibility but with limited performance software routers are only adopted for low-end devices. To improve the performance of software router, both horizontal and vertical scaling themes have been investigated. The former approach tries to build clusters with multiple nodes and the latter one tries to improve the performance of single node with additional processors like GPU (Graphics Processing Unit). This paper focuses on the latter scheme and consists of three topics.1) Forwarding lookup algorithm on GPU2) Kernel-space GPU computing environment3) GPU accelerated software router prototypeThis paper has verified the feasibility of accelerating software router with GPU through experiments.