A Future-Oriented Cache Management for Mobile Games

Yue Hu, Meng Wang, Yingfeng Chen, Changjie Fan

Abstract: As a common but critical module, cache management plays more and more important roles in modern videos games, especially for the games on mobiles whose hardware is weaker than the desktop computers. Current cache management policies typically resort to heuristics designed for the common access patterns, which cannot fully utilize the information in mobile games and may fail on the complex access patterns in it. In this paper, we propose a future-oriented cache management (FOCM) to cope with the cache management problem in mobile games' typical scenarios. FOCM leverages a dual Markov model to bridge the cache management's decision-making process and the behavioural process of the players in games, resulting in modelling the potential access requests in the future more accurately. Based on this model, we propose two cache management methods based on imitation learning and reinforcement learning, with each one suitable for different types of games. The experimental results in both the cache simulation environment and the industrial massively multiplayer online role-playing game (MMORPG) demonstrate the effectiveness of the proposed methods.