The algorithm mines periodically the use patterns of Web server with FP-tree method under the constraint of cache size, and then allocates documents of patterns in different back-end servers and forwards client requests to related servers according to their contents, so that the cluster system can achieve high disk usage and cache hit ratio.