Interaction with stable storage in the modern world isgenerally mediated by systems that fall roughly into oneof two categories: a filesystem or a database. Databasesassume as much as they can about the structure of thedata they store. The type of any given piece of datais known (e.g., an integer, an identifier, text, etc.), andthe relationships between data are well defined. Thedatabase is the all-knowing and exclusive arbiter of ac-cess to data.Unfortunately, if the user of the data wants more di-rect control over the data, a database is ill-suited. At thesame time, it is unwieldy to interact directly with stablestorage, so something light-weight in between a databaseand raw storage is needed. Filesystems have traditionallyplayed this role. They present a simple container abstrac-tion for data (a file) that is opaque to the system, and theyallow a simple organizational structure for those contain-ers (a hierarchical directory structure)