Rel is a free, open-source, true relational database management system with an advanced query language called Tutorial D. Intended for educational purposes and written in Java, it can run on any OS with a Java VM. It uses Oracle’s Berkeley DB as a storage engine, which in turn inherits some of its functionality form SQLite. The user can interact with the database via either the Tutorial D language or a visual query language implemented in the Rel application. While Rel currently does not support the entirety of the Tutorial D language, it has features for a large subset of it. Using the JDBC driver, it can be incorporated into any application that uses Java, as long as the application doesn’t specify SQL syntax.
CJ Date and Hugh Darwen’s The Third Manifesto outlines a set of programming language conventions and restrictions that lead to a truly relational database language that also incorporates benefits commonly associated with object-oriented languages. Tutorial D is a query language developed by Date and Darwen that meets those specifications, developed for education. Seeing the lack of complete open source Tutorial D language implementations, Dave Voorhis released the first version of the Rel database management system in 2004 in order to provide an implementation of Date & Darwen’s specification for instructional purposes.
The Rel DBMS supports two ways of executing queries. The first is by directly sending queries in Tutorial D, where the language interpreter will parse and process them. The second is by using the Rel application's visual query language. Queries made with this are converted into Tutorial D and are then fed to the language processor.
Multi-version Concurrency Control (MVCC) Two-Phase Locking (Deadlock Prevention)
Concurrency control is handled by the storage engine, so the concurrency control protocols in Rel are those that are implemented by BerkeleyDB -- namely, MVCC and 2-phase locking.
Nested Loop Join Sort-Merge Join
Currently, Rel does not have any optimizations for joins beyond the algorithms used by the storage engine. Thus, in BerkeleyDB, the possible join algorithms are those supported by SQLite, and even in SQLite, there is only support for sort-merge join over unique keys.
N-ary Storage Model (Row/Record)
Rel’s storage engine is a key-value store. Due to the fact that key-value pairs point to the memory location of an object as well as its size, there is no bound on the length of a data item.
Serializable Snapshot Isolation
Isolation levels are handled by the storage engine, so the isolation levels available to Rel are serializable (when BerkeleyDB uses pessimistic concurrency control) and snapshot isolation (when it uses MVCC).
Rel’s storage engine, BerkeleyDB, uses SQLite to execute queries. SQLite uses the Tuple-at-a-Time execution model, so BerkeleyDB inherits that model, and Rel inherits it from BerkeleyDB.
Since Rel uses the disk-oriented BerkeleyDB as a storage engine, it is also disk-oriented.
As Rel has Oracle's Berkeley DB as its storage engine, it also uses fuzzy checkpoints.
Foreign keys are not specifically implemented in Rel, since they are not a part of the official Tutorial D syntax. However, foreign key relationships are a more specific instance of the language's data constraint rules, so the Rel DBMS still supports foreign keys through its constraint system.
While the backend storage engine that Rel uses, Oracle’s BerkeleyDB, uses the Key/Value and Document/XML models, Rel’s data model is truly relational, as specified by Date & Darwen’s Tutorial D syntax.
Rel consists of two parts: a Rel database server and DBrowser, the client application. The role of DBrowser is to send Tutorial D expressions to the server, while the server itself does the actual query processing. Multiple clients can be connected to the same server, but the server itself is a single node, meaning the system is shared-everything.