Research Assistants: S. Goel
Sponsors: Army Research Laboratory (Software Technology Branch), AT&T, UNISYS, Purdue Research Foundation
Emerging "next generation" applications such as multimedia and engineering (CAD/CAE) require man aging databases that are more complex than those in applications such as banks and airline reservation sys tems. In O-Raid we are exploring the object-oriented approach as a means for overcoming limitations of current database management systems. These limitations typically include a lack of support for complex data structures and operations, integrity checking, and triggers.
O-Raid integrates the relational and object models. As a result it retains the simplicity of the relational model while providing the functionality of the object model. Objects, classes, and inheritance are supported together with a predicate-based relational query language. O-Raid extends the implementation of an existing distributed relational system called Raid.
We have extended the standard SQL query language (called SQL++) to support queries involving objects. Currently, we are conducting experiments on O-Raid to identify the overhead involved in executing SQL++ queries. This gives us an opportunity to assess the effort required to extend a relational system with objects. A graphical user interface called O-UI has been engineered which allows access and direct manipulation of O-Raid database relations and objects.
We are currently investigating several issues including indexing of relations containing objects, selective replication of parts of composite objects, and providing integrated access to a variety of data sources through a federated object.
The traditional scheme of replicating an object in its entirety, for increasing availability, would incur high performance and storage overheads when applied to composite objects such as multi-media documents and large software projects. We propose a new replication scheme, which allows replication of selected parts (su bobjects) of the composite object. Such selective composite object replication can tune the replication granularity to meet both the availability and performance requirements of distributed applications, and at the same time minimize storage costs.
We propose modeling existing related distributed data by a composite object, where each data source is treated as a fragment of the composite object. The creation of a composite object, referred to as a federated ob ject, is a pragmatic approach to data integration. Unlike in heterogeneous database systems, in this approach each data source is not required to have full database system capabilities. Federated objects are especially suit able for computer-supported-cooperative work (CSCW), where coordination and communication among a group of people is required, and the type of collaboration is dynamic in nature.