Virtualization presents a variety of trends in development, including both hypervisor-based and container-based classes. The former virtualization has the characteristic of well isolated and convenient operation, the latter one is lightweight and supply shortcut. With the rapid development of IT technology and the depth in application, complex applications require both types of virtualization to provide services. This paper proposes a two-stage approach to resource management, the first stage is to resolve physical resources monitoring, statistics, allocation of decision-making and isolation, allows a single physical resource has the ability of multiple virtualization resource abstraction; the second stage schedule for solving the resource needs of users and the underlying physical resources in place in order to achieve the object of increased physical resource utilization. While first stage scheduler also takes into account the difference in heterogeneous physical resources, and introduces a weighted DRF algorithm evaluation of heterogeneous physical effect on application performance. Based on the CloudSuite experiments, on the premise of guaranteed QOS parameters, the system resource utilization of the whole system is effectively promoted 20%.