面向可编程数据平面的网络程序合并优化方案  

Network Program Merge Optimization Scheme for Programmable Data Plane

作　　者：李炯 胡宇翔[1] 崔鹏帅 田乐[1] 董永吉[1] LI Jiong;HU Yuxiang;CUI Pengshuai;TIAN Le;DONG Yongji(Information Engineering University,Zhengzhou 450001,China)

机构地区：[1]信息工程大学,河南郑州450001

出　　处：《信息工程大学学报》2024年第6期682-688,共7页Journal of Information Engineering University

基　　金：国家重点研发计划(2022YFB2901501)。

摘　　要：目前网络程序合并生成的程序存在较长的流水线结构且单个流水线阶段内对资源的利用不充分,可能造成网络程序合并后无法完成部署,降低数据平面资源利用率。针对上述问题,提出面向程序合并过程的编译优化方案,该方案包含关键节点标记、关键节点的转换、最优结果筛选3部分,分别完成表依赖图中关键节点的寻找、关键节点前后依赖节点的转化、基于流水线长度和流水线宽度筛选最优程序,最后完成多个P4程序的合并和资源优化过程。实验结果表明,基于所提方法,在网络程序合并过程中可缩短程序流水线长度,并提高单个流水线内的资源利用率。相比传统网络程序合并过程,合并后的程序对内存的占用大约节约10%。In the current process of merging network functions,a simple addition or one-sided optimi-zation of function-specific resources is often employed,neglecting the logical interdependencies of net-work functions within the program.This results in a lengthy pipeline structure and underutilization of resources in individual pipeline stages for the ultimately merged network program,potentially imped-ing deployment and reducing data plane resource utilization.To address these issues,a compilation op-timization scheme for the program merging process is proposed.This scheme comprises three key com-ponents:key node labeling,key node transformation,and optimal result screening.These components respectively facilitate the identification of key nodes within the table dependency graph,the transfor-mation of nodes preceding and following key nodes,and the selection of optimal programs based on pipeline length and width.Ultimately,this scheme completes the merging and resource optimization of multiple P4 programs.Experimental results demonstrate that the proposed method shortens the pro-gram pipeline length and enhances resource utilization within individual pipelines during the network program merging process.Compared to traditional merging processes,the resulting merged program achieves approximately 10%reduction in memory usage.

关 键 词：可编程数据平面 增量式部署 合并网络程序 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]