Map-reduce-merge: simplified relational data processing on large clusters