Object oriented database (OODB) technology is becoming popular and is being used to support a large number of advanced applications. These applications require not only additional data modeling capabilities, but also advanced database design techniques and efficient query processing. Vertical class partitioning is a database design technique that increases the efficiency of query execution because it reduces the accesses to irrelevant instance variables. Structural join index hierarchy (SJIH) is an indexing technique that facilitates direct access to complex objects and their component objects. It saves a lot of disk accesses in query processing by reducing the accesses to irrelevant object instances....[ Read more ]

Object oriented database (OODB) technology is becoming popular and is being used to support a large number of advanced applications. These applications require not only additional data modeling capabilities, but also advanced database design techniques and efficient query processing. Vertical class partitioning is a database design technique that increases the efficiency of query execution because it reduces the accesses to irrelevant instance variables. Structural join index hierarchy (SJIH) is an indexing technique that facilitates direct access to complex objects and their component objects. It saves a lot of disk accesses in query processing by reducing the accesses to irrelevant object instances.

This thesis presents an empirical evaluation study of vertical class partitioning and structural join index hierarchy in object oriented databases. It extends the work of SJIH, where in each tuple in an index file only contains object identifier OIDs (i.e., OID-Based SJIH). A query may only be interested in a small part of data of objects, but executing the query by OID-Based SJJH needs to load whole data objects into main memory that results in a lot of irrelevant data accesses. To eliminate accessing irrelevant data, an Attribute-Based Structural join index hierarchy (Attribute-Based SJIH) is proposed in this thesis.

The experimental results show that applying vertical class partitioning in OOBDs can save a lot of disk accesses when compared with processing a query in unpartitioned OODBs. Fan-out of object, degree of sharing of object and constrained pair up of classes play a very important role in the process of choosing the best SJIH to support complex object retrieval in OODBs. The comparison of the experiment results in OID-Based SJIH and Attribute-Based SJIH show that Attribute-Based SJIH performs better than OID-Based SJIH.