Abstract: The data locality is divided into two levels. One is called the node data locality, which placing tasks on nodes that contain their input data. The other one is called the rack data locality, which placing tasks on nodes whose rack contains their input data. A new scheduling strategy called DDRF is proposed which combines the DRF and the delay. The DDRF is not only able to meet high locality but also achieve fairness. In the DDRF, the simulation results show the influence on the efficiency of jobs’ implement.