摘要
GPU可以显著提升一些网络功能的性能,但在GPU加速的网络功能虚拟化(Network Function Virtualization,NFV)系统中,由于网络功能需要以虚拟化方式独立开发和部署,其CPU-GPU处理流水线的CPU处理阶段会有较大的额外开销,使得网络功能GPU加速的效果不明显。为解决该问题,提出一个新的支持GPU加速的NFV系统框架。利用服务链中网络功能之间共享数据和流状态的特性,设计了共享式状态管理机制,以减少网络功能中重复性的协议栈处理和流状态管理开销,提升GPU加速的效果。对原型系统进行评估表明,相比于现有的系统框架,该框架能够显著地降低多种GPU加速的网络功能中CPU处理阶段的时间开销,并在常见的网络功能服务链上实现了高达2倍的吞吐量提升。
The GPU can significantly improve the performance of some of the network functions.But for GPU-accelerated network function virtualization(NFV)systems,network functions need to be developed and deployed independently in a virtualized manner,thus the GPU acceleration can be relatively inefficient due to high CPU overhead within the CPU-GPU processing pipeline.To solve this problem,we propose a novel GPU-accelerated NFV system framework.It utilized the characteristics of shared data and flow state between network functions in the service chain,and provided a shared state management mechanism to reduce the repetitive protocol stack processing and flow state management overhead in network functions and increase the efficiency of GPU acceleration.The experimental evaluation of the prototype system shows that,this framework is able to reduce the CPU stage overhead in GPU-accelerated network functions significantly and can achieve up to 2 times throughput improvement on a commonly used network function service chain,compared with the state-of-the art solution.
作者
郭良琛
张凯
Guo Liangchen;Zhang Kai(School of Software,Fudan University,Shanghai 201203,China;School of Computer Science,Fudan University,Shanghai 201203,China;Shanghai Key Laboratory of Data Science,Shanghai 200433,China)
出处
《计算机应用与软件》
北大核心
2022年第2期113-119,137,共8页
Computer Applications and Software
基金
国家自然科学基金项目(61802066)
国家重点研发计划项目(2018YFB1004404,2018YFB1402600)
上海市扬帆计划项目(18YF1401300)。
关键词
网络功能虚拟化
GPU
框架设计
Network function virtualization
GPU
Framework design
