Cache allocation optimisation of user relationship group based on reinforcement learning

Abstract

In order to alleviate network congestion and reduce request delay, device-to-device caching technology will be an important part of modern communication networks. People always browse the Internet for things they are interested in. However, different people have different points of interest. Therefore, how to choose a suitable cache node when people share content is a challenge. In this study, an optimal cache node selection algorithm based on virtual delay is proposed, where the multi-armed bandit model is used to obtain optimised cache decision based on the interest differences of users. Each candidate user may become a cache node, and the algorithm selects the user who minimises the overall delay as the node. When multiple candidate users exist, it is necessary to cooperate between multiple candidate users whose cache space is limited in order to maximise the cache hit rate. However, since each candidate user acts as a cache node has a defferent service efficiency for different requests from surrounding users, it is necessary to effectively distinguish cooperative cache users. This study proposes a master–slave node cooperative cache model based on the optimal cache node selection. The experimental results confirm that the proposed schemes achieve lower overall delay performance.