Online Learning Algorithms for Context-Aware Video Caching in D2D Edge Networks

With the emergence of various short video platforms such as TikTok and Instagram, coupled with the accelerated pace of people's lives, people are spending more time sharing and watching online videos than ever before, and they gradually turn their attention to short videos with short duration and novel content. Browsing and watching short videos by users with their energy-capacitated devices, such as phones and tablets, have become one of the main ways for users to entertain online. Timely response and high quality online video delivery are of the utmost importance to guarantee the quality of service (QoS) experienced by users. Caching videos at locations close to the video demanders can significantly improve the QoS by reducing the access delay of high quality videos. Together with an explosive growth of mobile devices and great demand for bandwidth, Device-to-Device (D2D) network is emerging as a promising technology to enable ultra-low latency communications, by allowing mobile devices to communicate with each other with or without the involvement of network infrastructures. Caching videos in D2D networks can further reduce video response delays of high quality videos, thereby improving the QoS experienced by users when watching short videos. However, how to cache the most appropriate videos at strategic mobile devices is crucial to satisfy the QoS requirements of users. Specifically, the QoS experienced by users depends on many intertwining factors from both users and the D2D network, such as videos characteristics, various users' demands for different videos, different communities of users, and energy levels of devices. Motivated by these facts, we investigate the video caching problems in a D2D network. The novelty of our study is to jointly consider the intertwining factors from both users and the D2D network when users access short videos. Specifically, we first formulate an optimization problem of video caching with the objective to minimize the average delay experienced by mobile devices, subject to the cache storage capacity and energy budget of each mobile device. We then propose an approximation algorithm with an approximation ratio for the offline video caching problem. We further devise an online learning algorithm for the online context-aware video caching problem. We finally conduct extensive experiments based on real datasets compared with existing similar studies. Experimental results demonstrate that our algorithms can achieve better average performance with confidence levels. For instance, our algorithms achieve 86% lower average delay experienced by users and 20% average energy consumption of each device, as well as 7% higher average hit ratio and 1.3 times more residual cache storage resource, than their counterparts.