People don't have to deal with data size and scalability issue? For me, I found it quite a pain having to make sure data fits the machine for in memory analytics. And handle OOM even when using parallel engine like Apache Spark.