面向异构ICN节点的副本选择算法研究  

A Study of Replica Selection Algorithm for Heterogeneous ICN Nodes

作　　者：高雷 朱小勇[1,2] GAO Lei;ZHU Xiaoyong(National Network New Media Engineering Research Center,Institute of Acoustics,Chinese Academy of Sciences,Beijing,100190,China;University of Chinese Academy of Sciences,Beijing,100049,China)

机构地区：[1]中国科学院声学研究所国家网络新媒体工程技术研究中心,北京100190 [2]中国科学院大学,北京100049

出　　处：《网络新媒体技术》2024年第4期26-34,共9页Network New Media Technology

基　　金：中国科学院声学研究所目标导向类项目:基于SEANET网络的分布式超算(项目编号:MBDX202114)。

摘　　要：信息中心网络(ICN)是一种革新式网络架构,打破了传统TCP/IP网络端到端传输的限制,提升内容分发效率。ICN构建全网规模的缓存系统,在网络内采用多副本冗余的方式缓存数据内容,以便用户就近获取。与传统互联网缓存系统不同,ICN的缓存呈现泛在化的特点,工作设备是网络基础设施,导致服务资源的异构性普遍存在。在这种环境下,选择适当的副本节点成为重要研究问题。本文首先通过M/M/1排队模型对异构ICN节点进行抽象建模和分析,然后将异构副本节点的选择建模成多臂老虎机问题,继而引入UCB1算法来探索并学习最优决策。仿真实验结果表明,该算法在提高缓存服务可靠性和缩短内容获取时延方面具有明显优势,算法使服务可靠性达到99.15%,将内容获取的平均时延最大缩短8.63%。Information Centric Networking(ICN)is an innovative network architecture that breaks the constraints of traditional TCP/IP end-to-end transmission and improves content distribution efficiency.ICN establishes a network-wide caching system with multiple content replica redundantly cached within the network for nearby access by users.Unlike traditional caching systems in the Internet,ICN’s caching is characterized as ubiquitous,with networking devices serving as the infrastructure,resulting in widespread heterogeneity of service resources.In such an environment,selecting appropriate replica nodes becomes a significant research challenge.This paper first conducts abstract modeling and analysis of heterogeneous ICN nodes with M/M/1 queueing model.The selection of heterogeneous replica is modeled as Multi-Armed Bandit(MAB)problem,and the UCB1 algorithm is introduced to explore and learn optimal decisions.Results of the simulation experiments demonstrate the algorithm’s significant advantages in terms of improving cache service reliability and reducing content retrieval latency.The algorithm achieves a service reliability of 99.15%and reduces the average content retrieval latency by a maximum of 8.63%.

关 键 词：信息中心网络 网内缓存 副本选择 M/M/1 排队模型 多臂老虎机问题 

分 类 号：TP393.02[自动化与计算机技术—计算机应用技术]