Implementing a filesystem in Hadoop