In response to the problems of real-time calculation and analysis broughted by massive data of future smart grid,we first built MapReduce frame based on a Hadoop cloud computing platform, demonstrated its powerful performance of data computing, and found many problems by an experiment in further improving computing efficiency for MapReduce - uneven scheduling task, offsetting data and the poor adaptability in heterogeneous environments, etc. Then we considered the drawbacks of the original scheduling program and gave an improved program for MapReduce architecture of mapping equilibrium data,assessing node performance,and proposed a dynamic matching scheduling algorithm（DMSA--Dynamic Matching Scheduling Algorithm）.Finally, we reduced the consumption of computing resources of the system, shortened the running time, significantly improved the performance of the cluster,and enhanced data locality through the experiment on the simulation platform. We proved the feasibility that this strategy can enhance the computing efficiency of MapReduce.