While MapRDB honors most of HBase's API, its still a different beast. (Both are in the same family, but still different species)
When you run a map/reduce job against MapRDB, what exactly is the access pattern?
Are you still making an RPC call to find the data?
How does data locality effect the query performance?
(There is more overhead in a read, so data locality becomes less of an issue)
Its been said that MapRDB is different from HBase in that it uses the file system. Is there any documentation or explanation beyond this?
I guess what I am looking at is a bit more detail on the data flow, something similar that you can find in Lars George's book on HBase.