With the advent of big data era, global storage is experiencing an explosive growth. Traditional storage systems have several drawbacks in storage performance, storage capacity, data security and device cost. In order to handle large amount of data, the storage technology for cloud computing platform has undergone rapid development in the recent years, and become an important tool to deal with big data. This paper analyzes the shortcomings of metadata management of certain distributed file system and proposes an adaptive metadata load balancing mechanism. First, a real-time server performance evaluation model is introduced. Next, a period adaptive mechanism based on change of server load is built. Finally, an adaptive load balancing algorithm based on server performance is proposed. Experimental results demonstrate the practicability, availability and stability of the new mechanism.