Abstract
NoSQL and MapReduce have taken roles in dealing big data. People seldom put these two methods together since they are suitable for different applications. NoSQL is designed for the Cloud. Users can use APIs supplied by NoSQL to query specific data from great amounts of data quickly. However, it would take minutes to hours to get data with pure NoSQL APIs in some cases. On the other hand, the MapReduce framework can parallel process data. With MapReduce, we may significantly shorten the processing time. In this work, we introduce a system architecture that automatic chooses APIs or MapReduce to retrieve big data from NoSQL. We propose a general prediction method for this new system architecture to hold both the advantages of APIs and MapReduce. We prove the feasibility of our design in the experiments.