A Multi-Agent Deep Reinforcement Learning based Cooperative Edge Data Caching Approach

Cloud based services are widely used in communication networks for the advantages of high computation and storage capability. However, requesting data from the cloud will trigger high communication delay. The data retrieving delay can be significantly reduced by providing storage resource at edge servers. Caching content at the network edges has been regarded as a promising technique in future mobile networks. However, the caching space size of the edge server is generally limited, thus the caching strategy of edge servers is required to be designed elaborately to optimize the content service. In this paper, this Cooperative Edge Data Caching (CEDC) problem is formulated as a Markov Decision Process (MDP) based multi-agent decision problem. We solve it by multi-agent deep reinforcement learning (MADRL) algorithm when the users’ preference is unknown and only the historical data requests can be observed. By introducing the concept of delay reduction rewards, the optimal caching decisions can be obtained based on the proposed algorithms. Finally, the proposed approach was evaluated on a real-world data set and the results demonstrate that it significantly advantages three representative approaches.