面向组播的动态虚拟网络功能放置算法  

Multicast-Oriented Dynamic Virtual Network Function Placement Algorithm

作　　者：邢焕来[1,2] 王心汉 宋富洪 赵博文 罗寿西[1,2] 戴朋林 李可 XING Huan-Lai;WANG Xin-Han;SONG Fu-Hong;ZHAO Bo-Wen;LUO Shou-Xi;DAI Peng-Lin;LI Ke(School of Computing and Artificial Intelligence,Southwest Jiaotong University,Chengdu 611756;Engineering Research Center of Sustainable Urban Intelligent Transportation,MoE,Chengdu 611756)

机构地区：[1]西南交通大学计算机与人工智能学院,成都611756 [2]可持续城市交通智能化教育部工程研究中心,成都611756

出　　处：《计算机学报》2023年第11期2322-2341,共20页Chinese Journal of Computers

基　　金：国家自然科学基金(No.62172342,No.62202392);四川省自然科学基金(No.2022NSFSC0568,No.2022NSFSC0944,No.2023NSFSC0459);中央高校基本科研业务费资助。

摘　　要：组播在支持日益增长的多媒体应用方面具有广阔的应用前景,面向组播的虚拟网络功能放置是网络功能虚拟化中不可避免的研究趋势.然而,对于该问题的大多数研究都聚焦于静态网络环境,难以应对网络中的各种资源随着时间动态变化,组播服务功能链(Service Function Chaining,SFC)请求动态到达的真实场景.本文提出一种基于组播SFC请求预测的足球联赛竞争算法,以Informer模型为基础,预测即将到达的组播SFC请求.基于足球联赛竞争的组播虚拟网络功能放置算法,设计多维个体编码策略,一次性求解所有活动组播组的SFC映射方案,提前部署预测的请求.针对预测结果与真实结果不一致的情况,提出一种由正向搜索与反向搜索组成的快速修复策略以完成对请求的快速响应.仿真结果表明,对比其它两种预测模型,Informer在组播SFC请求预测上取得了更低的均方误差与平均绝对误差.此外,与七种经典的启发式算法和深度强化学习算法相比,提出的算法在端到端时延和计算资源消耗方面达到更优性能的同时,取得了更低的组播SFC请求响应时间.Multicast,a potential technique to support ever-increasing multimedia applications,makes point-to-multipoint-oriented virtual network function(VNF)placement a promising research trend in net-work function virtualization(NFV).However,most existing research is hardly adapted to time-varying re-sources and dynamic multicast service function chaining(SFC)requests in the real-world network.This paper proposes a soccer league competition algorithm with multicast SFC request prediction(SLC-MSRP),which can foresee the incoming multicast SFC requests(MSRs)based on the Informer model.Based on multi-dimensional individual coding strategy,SLC-MSRP handles the SFC mapping of all active multicast groups simultaneously and deploys the predicted MSRs in advance.A fast repair strategy consisting of for-ward and backward search phases is developed to handle the differences between the predicted results and the actual arrival MSRs.Simulation results show that Informer achieves lower mean square error(MSE)and mean absolute error(MAE)values in the MSR prediction than two existing prediction models.Fur-thermore,the proposed algorithm gains lower MSR response time while achieving better performance in terms of end-to-end delay and computational resource consumption compared with seven state-of-the-art heuristic and deep reinforcement learning algorithms.

关 键 词：网络功能虚拟化 虚拟网络功能放置 组播SFC请求预测 足球联赛竞争算法 INFORMER 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]