Improving resource utility could increase the benefits of cloud computing service providers and decrease the price paid by users in the end. Therefore, how to scheduling tasks for resource is a key problem. Most of scheduling schemes proceeding nowadays have no QoS differentiation, which is necessary for cloud computing service operation. More than that, they are central architecture. The central node would be overloaded when the number of users is huge. And the central scheduler relies on messages to update the global status information.