Rick Stevens pointed me at this post describing Yahoo!'s work with Hadoop, the open source implementation of MapReduce. Of course, that is old news, but some navigation led me to this nice MapReduce cookbook for machine learning.
Many many years ago I wrote a book on parallel programming. My initial plan was to show how just about everything could be expressed using map reduce, but I couldn't quite see how to package things in a way way that made sense. (Probably I wanted to do it all via language constructs.) So instead I ended up covering a lot of other material instead. It's neat to see how with the right abstractions things turn out so nicely.