Abstract: Faced with the problem of explosive growth of data brought by the rapid development of the Internet of things industry, NoSQL can better handle massive data. In this paper, the typical MongoDB database in NoSQL is selected for research. The storage architecture of MongoDB is analyzed, and the auto-sharding mechanism of MongoDB is studied. Aiming at the shortage of current MongoDB load balancing algorithm, a load balancing strategy based on data block access frequency (DBAF) algorithm is proposed. The experiments show that the new load balancing strategy can effectively improve the concurrent read-write performance of the system.