The number of devices connected by the Internet of Things (IoT) has exceeded billions today. As IoT grows, so do the volumes of data it produces. Distributed data process (such as: MapReduce) performs better than a single central server by considering the big data set. Moreover, many IoT network spans widely in geographical areas, such as smart cities and supply chain management. Thus, the collected data can be processed in distributed way before transmission. This paper studies how to develop a MapReduce framework to process massive IoT data. Because in many cases IoT consumers desire more to get the meaningful knowledge than to build connections with multiple devices. The proposed framework should also fits well with the information-centric nature of IoT applications. As a result, our design (MR-IoT) is built upon a novel Information Centric Networking (ICN) architecture - NDN. It defines two schemes to execute MapReduce tasks on IoT: computational tree construction and computational task dissemination. A testbed is built on ndnSIM to verify the design and the result shows the defined schemes work correctly and the network traffic is significantly decreased.