Abstract: The cluster of Ceph cloud storage will have a greater performance differences in reading and writing different sizes of documents, when the underlying OSD nodes use different single file systems.To solve this problem, this paper proposes a hybrid file system approach in the Ceph cluster to optimize the performance of reading and writing.Through the use of analytic hierarchy process(AHP), this method considers the advantages of different file systems in reading and writing different size files, constructs a hierarchical model, gets the judgment matrix of each level and carries out the consistency test.Then, the normalized feature corresponding to the maximum eigenvalue of the judgment matrix is worked out to determine the proportion of the underlying OSD nodes using different file systems.By testing in the real environment, the result shows that this method can make up the shortcomings of the current Ceph cluster using a single file system, achieve the purpose of optimizing the speed performance in reading and writing different size of files.