I've been getting into couchdb and learning about map/reduce. I have read about it before but writing my own map and reduce functions and playing with simple datasets have been really eye-opening. There is one area that I'm still trying to grasp. CouchDB is built on map/reduce but it doesn't seem to use map/reduce in a distributed fashion. If i call a couchdb server process a "node", i can issue a query to a node and have it map/reduce the answer. I can start five nodes and they can keep in sync. I can send five queries and have each of the five nodes work on the answer to their question. What I thought the "big win" as far as map/reduce is, is to issue one query and have it divided amongst the five nodes to get the answer with some kind of linear speedup for that one answer.
Thanks in advance, Don
