Task scheduling is essentially an NP-completeness problem in cloud computing and the existing task scheduling strategies can’t fully meet its demands. In this paper, a feasible and flexible dynamic task scheduling scheme DGS is proposed, which dynamically allocates virtual resources to execute computing tasks and promptly completes the scheduling and execution process by using improved greedy strategy. The simulation platform CloudSim is expanded to realize the proposed scheme and the simulation results show that DGS can speed up the tasks’ completion time and improve the utilization of cloud resources to achieve load balance