Unclear - the company isn't really committed to it, so it has been relegated to the "long-term, low-priority" category. In other words, nothing is happening. :-(
Could change, and I hear that others may implement the concept given GP's inability to complete it.
It would be open source. The OMPI part of it is already in the OMPI developer's trunk - it's only the Hadoop integration that is missing.
You can run a MapReduce job just fine with OMPI as it sits - you just can't use the Apache MR classes. So if you want to write your own mapper/reducer, you can just execute it using mpirun with a few options that tell it to use a MR-like flow.
users mailing list