摘要
作为移动边缘计算(Mobile Edge Computing, MEC)的一项重要技术支撑,多边缘协作缓存的出现可更好满足终端智能应用的实时计算与数据存储需求进而提升用户体验.但是,多边缘协作缓存的性能通常受限于低效率的协作机制以及不合理的缓存资源配置策略.同时,如何在离散的用户特征分布与多样化的内容库之中寻找其潜在关联以提升缓存命中率仍是一个巨大的挑战.为了解决上述重要挑战,本文提出了一种新颖的基于联邦深度学习的多边缘协作缓存(Multi-edge Collaborative Caching with Federated deep learning, M2CF)方法.在M2CF中,首先设计了一种新型的多维缓存空间划分机制,对MEC节点的缓存空间进行感知优化,使得用户在分类区间可获得精准的内容推荐.接着,设计了一种基于VQ-VAE的内容流行度预测算法,解决了后验坍塌问题并提高了区间用户内容流行度预测的准确性.最后,设计了一种基于联邦深度学习的模型训练与缓存替换策略,通过聚合各MEC节点的本地模型以生成全局共享模型,进而更好适应优化后的不同缓存资源配置,提升多边缘协作缓存的命中率.基于MovieLens电影评分真实数据集,本文在测试床上展开了大量对比实验对所提出的M2CF方法进行了全面的评估.实验结果表明,M2CF与其他缓存方法对比展现出了更优秀的缓存性能与时效性能,且可以适应更为复杂的多边缘场景.
As an important technical support for Mobile Edge Computing(MEC),the emerging multi-edge collaborative caching can better meet the real-time computing and data storage requirements of end intelligent applications and improve user experience.However,the performance of the multi-edge collaborative caching is commonly limited by inefficient collaborative mechanisms and unreasonable strategies of cache resource configuration.Meanwhile,it is still challenging to find the potential relations the discrete distributions of user features and diverse content libraries for improving the cache hit rate.To address these important issues,this paper proposes a novel Multi-edge Collaborative Caching with Federated deep learning(M2CF)method.In M2CF,a new mechanism of multi-dimensional cache space partitioning is first designed to perceptually optimize the cache space of MEC nodes,which enables users to obtain accurate content recommendations in the classification interval.Next,a VQ-VAE-based content popularity prediction algorithm is designed to solve the problem of posterior collapse and improve the content popularity prediction accuracy of interval users.Finally,a model training and cache replacement strategy is designed based on federated deep learning.Through aggregating the local models of MEC nodes to generate a global shared model,the designed strategy can better adapt to the optimized configurations of different cache resources and improve the hit rate of multi-edge collaborative caching.Using the real-world movie-rating datasets of MovieLens,extensive experiments were carried out on the test bed to comprehensively evaluate the proposed M2CF method.The experimental results show that M2CF has better cache performance and timeliness compared with other caching methods,and can adapt to more complex multi-edge scenes.
作者
梁杰
郑家瑜
陈哲毅
于正欣
苗旺
LIANG Jie;ZHENG Jiayu;CHEN Zheyi;YU Zhengxin;MIAO Wang(College of Computer and Data Science,Fuzhou University,Fuzhou 350116,China;Key Laboratory of Spatial Data Mining&Information Sharing,Ministry of Education,Fuzhou 350002,China;Fujian Provincial Key Laboratory of Networking Computing and Intelligent Information Processing,Fuzhou 350116,China;School of Computing and Communications,Lancaster University,Lancaster LA14YW,UK;School of Engineering,Computing and Mathematics,University of Plymouth,Plymouth,PL48AA,UK)
出处
《小型微型计算机系统》
CSCD
北大核心
2024年第12期2994-3001,共8页
Journal of Chinese Computer Systems
基金
国家自然科学基金项目(62202103)资助
中央引导地方科技发展资金项目(2022L3004)资助
福建省财政厅科研专项经费项目(83021094)资助
福厦泉国家自主创新示范区协同创新平台项目(2022FX5)资助。
关键词
移动边缘计算
多边缘协作缓存
联邦深度学习
多维缓存空间划分
内容流行度预测
mobile edge computing
multi-edge collaborative caching
federated deep learning
multi-dimensional cache space partitioning
content popularity prediction
