Thanks to the development of virtualization technology, the private and public cloud data center has more and more appeared in enterprises, schools and research institutions. Relative to the physical machine, the virtual machine has better mobility,scalability,and relatively cheap to buy and maintenance, so more and more small and medium-sized entrepreneurs tend to buy virtual machine deployment services. So for cloud service providers, how to allocate physical resources to the VMs in one cloud environment and maximize the utilization of hardware resources pool meeting SLA is becoming more and more important. In this paper, we analyzed the key parameters that affect the VMs' application performance, we used a method of SVD + non-linear model in the application performance modeling and a good results were obtained in the experiment that makes the average prediction error can reach about 12%.