Abstract: With the development of Internet of Things, the sensing data gathered by large amounts of sensors shows the massive, continuous and probabilistic characteristics.In order to satisfy the requirements of real-time processing of such sensing data, the share-nothing architecture of cloud computing is adopted to support the parallel sensing data processing with extendibility, and a parallel computing model and corresponding programing interface for real-time sensing data processing is proposed by extending the MapReduce.Furthermore, a decentralized distributed architecture and cluster management method are designed to implement the parallel computing model.The experiment shows that the system has good scalability and the processing performance increases in linear progression as the number of server increases.