Dealing with Hadoop’s small files problem