The Qddb Database Suite

In the simple case, Qddb uses a single set of tuple trees instead of flat tables. A tuple tree is simply the collection of prejoined rows from the relational tables in a one-many relationship. That is, each record represents some entity, and each entity can have an arbitrary number of values for some subset of the fields. More complex cases (such as many-many relationships) can be handled with a set of Qddb schemas.

We can use standard relational tables to model any Qddb schema. For example, our Properties schema requires four tables: one table for the address information and one table for each expandable field. The tables require a plethora of link information for the purpose of joining the rows.

It is the complexity of managing these tables that prevents normal users from building their own databases with standard relational tools. Even programmers avoid fields that can have multiple values because a separate table is required for each such field. With Qddb, users and programmers can allow multiple values for any field by simply appending an asterisk to the field's definition in the schema.

The relational model requires that field values be atomic. That is, when searching, field values cannot be broken into smaller searchable components. By default, Qddb disobeys this rule by breaking field values of type string at the separators. Most users do not care about the relational model, they simply want to do their work. Sometimes this work involves searching the database for a particular record. By relaxing this restriction, Qddb can search for words in textual fields such as comments or abstracts. For string fields where atomicity is important, you can specify an empty list of separators for that particular field.

Qddb uses several interesting storage techniques. If you look at the contents of the files in your database directory, you will see that all Qddb data and index files are stored as readable text. The Database file contains the field values for each record in the stable part of the database. Empty fields require no storage and are omitted from the Database file.

If you browse the Database file, you will also notice that each record is stored contiguously. Qddb's records are prejoined rows from the relational tables defined in the schema. When you perform a search, the matching tuple trees in the Database file are expanded into the equivalent relational rows for viewing purposes.

Qddb uses inverted indices for searching. When you perform a search, the criteria are quickly translated into the file offsets for the corresponding tuple trees. Each matching tuple tree can then be read from the disk with one disk read per tuple tree.

Since Qddb currently uses inverted indices for indexing, you should periodically reindex the database so Qddb can retain its speed. Only changes and additions are stored in a secondary location and can be reindexed on the fly. The qstall command is used for this purpose:

qstall Properties

After stabilization, all changes, additions, and deletions are committed to the Database file.

History and More Information

Resources