SUMMARY
Data centers are created as distributed information systems for massive data storage and processing. The structure of a data center determines the way that its inner servers, links and switches are interconnected. Several hierarchical structures have been proposed to improve the topological performance of data centers. By using recursively defined topologies, these novel structures can well support general applications and services with high scalability and reliability. However, these structures ignore the details of some specific applications running on data centers, such as MapReduce, a well-known distributed data processing application. The communication and control mechanisms for performing MapReduce on the traditional structure cannot be employed on the hierarchical structures. In this paper, we propose a methodology for performing MapReduce on data centers with hierarchical structures. Our methodology is based on the distributed hash table (DHT), an efficient data retrieval approach on distributed systems. We utilize the advantages of DHT, including decentralization, fault tolerance and scalability, to address the main problems that face hierarchical data centers in supporting MapReduce. Comprehensive evaluation demonstrates the feasibility and excellent performance of our methodology.