The documents are stored on disk (in the file system) using a simple hash. The object id is used to create what you see on disk. So an object with the id f5da36d9-bcba-421e-9932-317b3ae2ed4e will be hashed to the folder f5/da (assuming you have the default # of folder levels of 2). The object id is stored in the relational database and used to find/access the object. This approach is similar to almost every scalable document management/archive system I have ever seen. It is a highly scalable, fast architecture and allows for single object store (meaning if two binaries map to the same hash and are the same file then only one copy is stored).