Edge Caching Based on Deep Reinforcement Learning
Farnaz Niknia, Ping Wang, Aakash Agarwal, Zixu Wang · 2023
The use of mobile apps has caused a lot of data to be transmitted repeatedly, which has put a strain on the networks. One solution to this problem is caching, which stores data closer to the user to reduce data transfer and delay. Typically, when designing a caching policy, only a limited set of file features, such as freshness and popularity, are taken into account in the majority of prior studies. Nevertheless, it is crucial to incorporate factors like importance and size since highly valuable files may hold more significance compared to freely accessible ones. In this paper, we modeled caching problem using Semi-Markov Decision Process that considers file features such as popularity, lifetime, size, and importance. Then, we used a Reinforcement Learning algorithm called Double Deep Q-Learning to solve the Semi-Markov Decision Process. Simulation results show that the proposed method outperforms an existing caching method in terms of cache hit rate and total utility in various settings. This method is the first to comprehensively consider all file features, making it more practical for real-world scenarios.