The invention provides a Map Reduce task data-center-across scheduling system and method based on a master-slave framework. The system and method are characterized in that a Map Reduce task overall task scheduling center is built and is responsible for managing resource managers of data centers and receiving Map Reduce task requests submitted by a client side Client, and the data centers meeting the requirement are selected according to a preset scheduling algorithm; the resource managers of the data centers synchronize the states of the centers and task execution information to the overall task scheduling center at regular time. According to the system and method, data-center-across scheduling of a Map Reduce task is achieved, a uniform entrance is provided for the Map Reduce task across the data centers, and the data and computing resource sharing of the data centers is effectively achieved.