iMapReduce: A Distributed Computing Framework for Iterative Computation

Relational data are pervasive in many applications such as data mining or social network analysis. These relational data are typically massive containing at least millions or hundreds of millions of relations. This poses demand for the design of distributed computing frameworks for processing these data on a large cluster. MapReduce is an example of such a framework. However, many relational data based applications typically require parsing the relational data iteratively and need to operate on these data through many iterations. MapReduce lacks built-in support for the iterative process. This paper presents iMapReduce, a framework that supports iterative processing. iMapReduce allows users to specify the iterative operations with map and reduce functions, while supporting the iterative processing automatically without the need of users' involvement. More importantly, iMapReduce significantly improves the performance of iterative algorithms by (1) reducing the overhead of creating a new task in every iteration, (2) eliminating the shuffling of the static data in the shuffle stage of MapReduce, and (3) allowing asynchronous execution of each iteration, {it i.e.,} an iteration can start before all tasks of a previous iteration have finished. We implement iMapReduce based on Apache Hadoop, and show that iMapReduce can achieve a factor of 1.2 to 5 speedup over those implemented on MapReduce for well-known iterative algorithms.