I like Spark over Hadoop just from an interface point of view, particularly the ability to just start up a (Scala) shell and start playing around. Hadoop can be very effective, but even getting "hello world" to run requires an intimidating array of setup.