running spark application on hadoop