The iMapReduce project started at University of Massachusetts Amherst in 2010. iMapReduce is a modified Hadoop MapReduce Framework for iterative processing. iMapReduce allows users to specify the iterative computation with the separated map and reduce functions, and provides support of automatic iterative processing. More importantly, it improves performance by
Even though it is not true that every iterative algorithm can benefit from all these three features, most of graph-based iterative algorithms are quite suitable in iMapReduce, e.g. shortest path and pagerank. The other iterative algorithms, e.g. NMF, kmeans, at least, can benefit from the first feature. So we can see different performance improvement for different applications.
For more information, please take a look at our iMapReduce paper in DataCloud'2011.
This project is a prototype implementation of the iMapReduce idea. The prototype isbased on Hadoop 0.19.2 and HOP. It is better used for research perspective, but we don't recommend to use it in production. Of course, we welcome any feedback on iMapReduce.