Efficient and scalable metadata management system is the key to improve the overall performance of distributed storage systems. Traditional metadata allocation strategies result in load imbalance of metadata, and in the case of multi-process resource preemption, there exists low efficiency in response to user requests, limitation on the number of files stored and etc. The system cannot meet the demands of high-concurrency, low-latency. To solve these problems, a distributed metadata management system based on consistent hash and directory tree, is presented in this paper. It introduces a load balance algorithm for metadata migration which has low cost and fine granularity. Experiments show that this system can achieve the load balance of metadata, obtain low respond latency and effective distribution of metadata. and improve the scalability and availability of distributed systems.