Abstract：It's far less effective for the stateless accelerator to accelerate the stateful network function.In order to solve the problem,this paper presents a programmable hardware-based stateful network function acceleration architecture which is called Stateful Function Processing Acceleration (SFPA) architecture.Providing the Stateful Processing Unit (SPU) to the data plane,SFPA can offload the data processing task to the data plane.In addition,SFPA can allocate the acceleration resources to multiple VNFs independently,decrease hardware cost and improve the flexibility of the acceleration architecture with the resource allocation optimization algorithm.Results of the experiments which are based on the NetFPGA-10G platform show that the throughput of VNF is 2.9 times faster than that of DPDK,and 1.7 times faster than that of stateless hardware accelerator in the SFPA.The optimal rate of resource allocation optimization algorithm is up to 41.9%.
[1] 徐雷.网络功能虚拟化技术与应用[M].北京:人民邮电出版社,2016.
[2] Kourtis M A,Xilouris G,Riccobene V,et al.Enhancing VNF performance by exploiting SR-IOV and DPDK packet processing acceleration[A].Proceedings of IEEE Conference on Network Function Virtualization and Software Defined Network (NFV-SDN)[C].San Francisco:IEEE,2015.74-78.
[3] Bronstein Z,Roch E,Xia J,et al.Uniform handling and abstraction of NFV hardware accelerators[J].IEEE Network,2015,29(3):22-29.
[4] Ge X,Liu Y,Du D H C,et al.OpenANFV:Accelerating network function virtualization with a consolidated framework in openstack[J].ACM SIGCOMM Computer Communication Review,2015,44(4):353-354.
[5] Bi J,Zhu S,Sun C,et al.Supporting virtualized network functions with stateful data plane abstraction[J].IEEE Network,2016,30(3):40-45.
[6] Sun C,Bi J,Hu H,et al.NeSMA:Enabling network-level state-aware applications in SDN[A].Proceedings of IEEE 24th International Conference on Network Protocols (ICNP)[C].Singapore:IEEE,2016.1-7.
[7] Kablan M,Caldwell B,Han R,et al.Stateless network functions[A].Proceedings of the 2015 ACM SIGCOMM Workshop on Hot Topics in Middleboxes and Network Function Virtualization[C].London:ACM,2015.49-54.
[8] 马宜科,常晓涛,范东睿,等.混合体系结构中有状态硬件加速器的优化[J].计算机学报,2011,34(7):1314-1322. MA Yi-ke,CHANG Xiao-tao,FAN Dong-rui,et al.Optimization of stateful acceleration in hybrid architectures[J].Chinese Journal of Computers,2011,34(7):1314-1322.(in Chinese)
[9] Gibb G,Varghese G,Horowitz M,et al.Design principles for packet parsers[A].Proceedings of the Ninth ACM/IEEE Symposium on Architectures for Networking and Communications Systems[C].San Jose:IEEE Press,2013.13-24.
[10] 段通,兰巨龙,胡宇翔,等.一种支持网络功能演进的可重构数据平面[J].电子学报,2016,44(7):1721-1727. DUAN Tong,LAN Ju-long,HU Yu-xiang,et al.A reconfigurable dataplane enabling network function evolution[J].Acta Electronica Sinica,2016,44(7):1721-1727.(in Chinese)
[11] Jose L,Yan L,Varghese G,et al.Compiling packet programs to reconfigurable switches[A].Proceedings of Usenix Conference on Networked Systems Design and Implementation[C].Oakland:USENIX Association,2015.103-115.
[12] Ruiz M,Ramos J,Sutter G,et al.Accurate and affordable packet-train testing systems for multi-gigabit-per-second networks[J].IEEE Communications Magazine,2016,54(3):80-87.
[13] Duggisetty P.Design and Implementation of a High Performance Network Processor with Dynamic Workload Management[D].Amherst Massachusetts:University of Massachusettes,2015.
[14] Bi H,Wang Z H.DPDK-based Improvement of Packet Forwarding[OL].https://www.itm-conferences.org/articles/itmconf/pdf/2016/02/itmconf_ita2016_01009.pdf,2016-11-21/2017-03-06.
2018, Vol. 46 
