I have been reading a lot about MapReduce and it's open source counter part - Hadoop. So, I decided to read up on the official paper and write a small simulation MapReduce framework that just simply does the Map and Reduce operations.
This simulator does not have any of the fancy features that Google's MapReduce is designed with. ie. Fault tolerance, failover, etc.... However, it does allow you to run the mapper task as multiple threads. If you want to get an idea of how to write programs using MapReduce without breaking a sweat, please go ahead and experiment!
The sample program given on the last page in the paper above can be run almost unmodified with the framework I present. You can set the number of threads to use using the spec.set_threads(XX) function.
You can download this simulator framework along with 2 test programs,
WordCount and Grep from here.