Interesting article and integration. Combining the power and elegance of the Mathematica language with the distributed processing capabilities of Hadoop could be very intriguing. However:
“Notice that a Mathematica kernel is required on each slave node running MapReduce tasks.”
Unless Wolfram comes up with some attractive license options, I suspect this will be very limiting. The rapid adoption and growth of ecosystems around free platforms such as Hadoop or R will be hard to keep up with, especially among students and big data entrepreneurs.
What are some of the largest M + Hadoop clusters today?
What volume license terms are we looking at, say for 1000 nodes?