Design of an Intelligent Memory Reclamation Service on Android

Android has shown its success in the current smartphone market. However, the memory management issue is not extensively addressed in Android because it employs a naive LRU-based reclamation scheme to solve the problem of lacking enough free memory while launching large applications. Therefore, the user may perceive a long loading time for such situations. In addition, the LRU-based scheme may clumsily kill some activities to be re-invoked in a very short time, and thus result in future unnecessary reclamation processes. In this paper, we describe the design of an intelligent memory reclamation service that employs the Markov Decision Process (MDP) model to learn the user behavior patterns and perform the reclamation process. We have conducted experiments based on artificially generated activities. The experimental results demonstrate that the proposed intelligent memory reclamation service can effectively learn the user behavior patterns and have 22.1% loading latency improvement to the default LRU-based reclamation scheme on average. The promising results suggest practical applications in the future.