SaaS为软件发展的最新趋势，它需要采用Web集群提供高性能的服务。如何在绝对保证各用户级别多种性能指标的前提下使集群的能耗最低是运营商们急需解决的问题。.首先提出一种服务节点支持CPU频率调整的多性能指标QoS保证方案，它通过合理地设计节点工作参数，并采用基于广义预测控制和T-S模糊模型的MIMO控制方案对其值进行测定，使得当节点工作在测定的工作参数值上时请求平均响应时间和丢弃率都能得到保证。然后采用经证明合理的规划变量定义方式，将集群高能效QoS保证的优化部署问题（包括各节点的开关，各开启节点的CPU频率及其服务的用户级别）描述成混合整数规划问题，由于变量数目大幅度减少，使得在线求解该问题成为可能。最后提出一种求解该问题的改进差分进化算法，它能较快地收敛到全局最优可行解。.项目成果不仅具有很好的应用前景，并对其他模式的云计算、网格计算等相关领域的QoS和节能研究也有很好的参考价值。

SaaS(Software as a Service) is the latest trend of software development, and it needs to adopt web cluster to supply high performance service. How to minimize the energy consumption on the premise of absolutely guaranteeing every user-level's multiple performance-indexes is an urgent problem for SaaS providers..First, a multiple performance-indexes absolute QoS guarantee scheme for single server node is proposed, which supports CPU frequency adjustment. The scheme properly selects and designs server node's working parameters, and adopts an MIMO control scheme based on Generalized Predictive Control and T-S fuzzy model to determine the working parameters' values, so that the requests' average response time and drop rate of a server node can be guaranteed simultaneously when the node is working under the predetermined working parameters' values. Then, by adopting a programming variables definition method which is proved to be reasonable, the cluster's optimization deployment problem (including every node's on/off state, every working node's CPU frequency, and the user-level every working node serves) of high energy-efficient QoS guarantee is described as a Mixed Integer Programming problem. The substantial reduction of programming variables' number makes the online solution of the programming problem feasible. Finally, an improved Differential Evolution algorithm is proposed to solve the above programming problem. The algorithm can converge to the global optimal feasible solution quickly..The research achievement of the project not only has very good application prospect, but also has very good reference value for QoS and energy-saving research in cloud computing of other modes, grid computing, and other related areas.