题目/Title：IP Routing Processing with Graphic Processors

作者/Author：穆帅,张昕雅,张乃燃,鲁嘉鑫,邓仰东,张舒
                        Shuai Mu,Xinya Zhang,Nairan Zhang,Jiaxin Lu,Yandong Deng,Shu Zhang

会议/Conference：DATE 2010

地点/Location：Dresden

年份/Issue Date：2010.8-12 March

页码/pages：pp. 93 - 98

摘要/Abstract：
Throughput and programmability have always beenthe central, but generally conflicting concerns for modern IProuter designs. Current high performance routers depend onproprietary hardware solutions, which make it difficult to adaptto ever-changing network protocols. On the other hand, softwarerouters offer the best flexibility and programmability, but couldonly achieve a throughput one order of magnitude lower. ModernGPUs are offering significant computing power, and its dataparallelcomputing model well matches the typical patterns ofpacket processing on routers. Accordingly, in this research weinvestigate the potential of CUDA-enabled GPUs for IP routingapplications. As a first step toward exploring the architecture ofa GPU based software router, we developed GPU solutions for aseries of core IP routing applications such as IP routing tablelookup and pattern match. For the deep packet inspectionapplication, we implemented both a Bloom-filter based stringmatching algorithm and a finite automata based regularexpression matching algorithm. A GPU based routing tablelookup solution is also proposed in this work. Experimentalresults proved that GPU could accelerate the routing processingby one order of magnitude. Our work suggests that, with properarchitectural modifications, GPU based software routers coulddeliver significant higher throughput than previous CPU basedsolutions.