A Highly Reliable Data Storage Method for Power Data Centers with Hyper-Converged Architectures

The data center suffers from poor reliability and unbalanced load. This is caused by the selection of single-round storage nodes that are too close to each other. This paper proposes a data storage method based on deep reinforcement learning, which includes two parts: the node reliability judgment module and the data placement module. Firstly, we use multiple neural networks to build a node reliability judgment module to evaluate the reliability degree of server nodes. Based on this, a distributed parallel data storage method is proposed to select server nodes. The experimental results show that the data storage method in this paper has high reliability and can achieve load balancing in the data center.