Resource Management in Cloud Based on Deep Reinforcement Learning

Abstract

In order to provide users timely and reliable services, resources and facilities in cloud data centers are often more than actual needs, which leads to low resource utilization and high operating costs. However, controling the quantity of resources and facilities may cause low throughput. Therefore, many researches are devoted to increasing resource utilization, improving jobs throughput and reducing operating costs through effective cloud resource management. The cloud resource management in cloud data center can be regarded as a multi-resource demand job allocation problem. This paper proposed an allocation algorithm of jobs based on deep reinforcement learning. The algorithm allocates waiting jobs to clusters respectively, so as to optimize throughput and improve resource utilization. The algorithm is implemented in cloud data center simulation model CloudSim. The experimental results show that the method based on Deep Reinforcement Learning is better than traditional method in multi-resource job allocation problem.