The increase of requests often results in performance degradation of the web server cluster. The quality of service (QoS) expected by users can't be guaranteed. This is the problem that service-level web server cluster must cope with. In this paper, two strategies are proposed including classification of the user requests and request admission control, the different quality of service can be reached and the QoS of service-level users is guaranteed. The simulation experiments proved, the load balance of the system is improved and average response time of the system is decreased.