Abstract: With the further expansion of Big Data and Big Data Centre, new extensibility requirements are proposed on the Hadoop Platform. In the base of systematically analysis constraints on performance and extensibility of Hadoop Platform, this paper presents a multi-level Hadoop Platform which has a structure of Global-Region-Node after being divided. In this platform, Master Node is responsible for metadata management and task development on the global level, and regional management node takes charge of metadata management and task development on the regional level, which improve both scalability and fault tolerance of Hadoop Platform. Experiments show that this platform can effectively promote the scalability of Hadoop and has certain feasibility.