Nowadays most of cloud data centers deploy high available system in order to provide continuous services, so it’s very important for a high available cluster to detect the node failure (physical machine failure) accurately and timely in a low bandwidth occupation way. However, compared to the traditional cluster environment, the scale of cloud data center increases rapidly with the use of virtualization, so traditional node failure detection models have already faced several new problems. In this paper, we present a three roles and two layers node failure detection model, named as Smart Ring, which fits cloud data center well and strikes a balance between accuracy, instantaneity and bandwidth occupation. It can simultaneously detect the status of physical machines and virtual machines and deal well with multiple nodes failure and network partition. Our experiment results show that Smart Ring has a better performance than most existing models.