The invention discloses a Hadoop optimization method based on asynchronous starting. According to the method, the operation is changed from serial execution into partial parallel execution, so that the whole operation executing process is optimized, the iteration operation speed is greatly accelerated, and the executing efficiency is effectively improved. The method has the advantages that bottom layer codes do not need to be modified, the use is convenient, in addition, the cluster utilization rate can be improved, and the bottle neck of a memory space does not exist. The invention is used as the Hadoop optimization method based on asynchronous starting, and can be widely applied to a Hadoop architecture technology.