Servlet cache can effectively improve the throughput of Servlet container and reduce the response time experienced by the users. But the cache effect is dependent on the hit rate determined by the cache replacement algorithms. Servlets represent some business functions, so mining the business association among Servlets can improve the hit rate of cache replacement algorithms which in turn exhances the performance of Servlet container consequently. However existing literatures such as LRU (least recently used), LFU (least frequently used), GDSF (greedy dual size frequency) rarely take into account the relationships between the Servlets. This paper denotes the business associations as sequential patterns of Servlet container, and presents a k-steps transfer probability graph to denote the access sequential patterns of Servlet container and designs a sequential patterns discovery algorithm KCTPG_Discovery. Two cache replacement algorithms KP-LRU and KP-GDSF are introduced based on the research of the sequential patterns of the Servlets. Comparing with the traditional algorithms such as LRU and GDSF, the experimental results confirm that the hit radio of the cache can be enhanced by using the above algorithms, the two algorithms effectively improve the performance of Servlet container.