Object Oriented databases

The object oriented databases originated from well-known limitations of the Relational DBMS which was too simple for CAD/CAM, GIS/CASE since there was limited datatypes and no UDTs at the time.  Among the other limitations were that RDBMS does not allow the concept of generics or generalization and aggregation. And there is a mis-match between DB language and application languages. There was no support for design transactions, versions and schema evolutions. Hence the approach to extended relational DBs ( or object-relational model ) with support for user defined data types, stored procedures or rules, transitive closures, and generalization.

Fundamentally, an object oriented database implements a conceptual object data model with objects and their identifiers. This model is all about encapsulation, classes, inheritance, overloading and overriding, complex values and types. Therefore it differs from a entity relationship model where there are explicit relations defined and schemas. The object model extends the relationships to add support for subclass and instances, composition hierarchy and checking out / checking in of objects etc. Integrity constraints are also extended in object databases with referential integrity constraints and uniqueness constraints and siblings or sub-class relationships.  Support for composite objects with references to other objects with exclusive or shared parts is also available.

Similarly, the query language and optimization has been extended. The queries look very similar to those in programming languages. Expression with support for All or Any are also made available. The issues with query optimization in the object-relational model include user defined methods (hard to optimize), class hierarchy based queries ( requiring new strategies), multi-valued attribute ( needs new cost models ) and dual buffering ( separately process objects in the buffer / on the disk).

Locking is at the granularity of objects (intention locks) on large complex objects. Transactions share objects in memory buffers and recovery needs to handle memory buffers.

An object database provides a new model closer to application development and requires refinements on the relational model. When objects are flattened, they lend themselves to storage in a relational model.