Drill Down: Beyond the Relational Database

Beyond the Relational Database

In 1970, E.F. Codd laid out the basics of relational database systems(RDBMS). As most people in the field know, the main tenet of his model proposed theabsolute separation of the logical view and the physical view of data. The logical view ofdata in a relational database is set-oriented and the relational set is structured as atable. The items are the rows of the tables and the columns are fields for each item. Eachcolumn has a distinct name; rows do not have names. A database consists of one or moretables plus a table or catalogue describing the database. Codd also described a datadefinition language and data manipulation language that allowed users to work with thedatabase. In essence, people could show relationships between items in different tableseither by joining the tables or creating new tables.

In Codd's model, both data storage and data processing can be distributed across theenterprise. Moreover, combinations of data can be created by drawing from informationstored in many locations while access to the aggregate information can be restricted. Forexample, employee information can be gathered from a network of branch offices, but viewedin total only at the headquarters. In the past 30 years, it has served as the basis of a$4 billion industry, which is currently the dominant model for database development fromthe desktop to the enterprise.

Despite the success of the relational database model, some observers think that it istime to move to new models. For instance, Andrew Wade, founder and Vice President ofObjectivity, Inc. (Mountain View, Calif.), a vendor of object-oriented databasetechnology, argues that relational technology is starting to show signs of age. Forapplications beyond those that can use simple tabular data models, short, simpleoperations and centralized data storage and access, relational databases hit what Wadecalls the "relational wall." When users want to use a table-based databasemanagement system for non-tabular data, or central-server architecture for a distributedapplication, or a non-scaleable approach for a project that needs to be scaleable,complexity grows. And as complexity increases, performance decreases.

Objectivity markets an object-oriented database management system (ODMBS). Instead oftables and operations, such as select and join, ODMS technology relies on objects thatencapsulate data and methods. Objects then interact with each other in ways the userdefines.

The difference between a relational and object-oriented database can be understood thisway: If all the information about a customer is removed from a table in a relationaldatabase, that customer, for all intents and purposes, ceases to exist in the database.The data is all there is. On the other hand, if the customer has been captured in anobject, the customer can continue to have an identity in the database, even if there is noparticular information about the customer. All information does not have to be based onthe most primitive information (i.e., the data fields in a record).

The move to object-oriented technology holds other advantages, as well, Wade argues.Let's say you want store a car in a garage. In a relational database, the car isrepresented by all the fields in the record "car." Each action on"car" involves all those fields. In an object-oriented approach, there can be anobject called "car" independent of the related information. Since every activityinvolving "car" only requires the use of one object, DBMS performance andintegrity can be significantly improved.

If the move to ODBMS can be seen as a big step, some do not believe it is big enough. Asmall research, development and consulting company in Crofton, Md., called AdvancedSystems Technology Corp. (ASTEC) is promoting what it calls a "semantic"information model for database technology, an approach that has been pioneered in the U.S.Air Force and other government agencies. (ASTEC also markets at data repository calledCaMERA, which stands for Culture-adaptive Mechanism for Expression, Reflection andAnalysis of Systems.) As Codd saw it, says Hassan Sayani, ASTEC's President, "if youwanted to perform mathematical functions on data, you needed a simple structure." SoCodd chose a table format with binary relationships. That is, each piece of data has onerelationship with any other piece of data.

Unfortunately, Sayani notes, the tables and binary relationship present severelimitations for handling complex information problems. Tables and binary relationships donot reflect the way people actually see the world, or the way the world works.

A semantic information model allows all the roles and relationships involved in thatscenario to be modeled and expressed as easily as they are in English, Sayani says. ODBMSshould be seen as a subset of a semantic DBMS. "In an ODBMS, all objects are data andmethods. So it is like all you have are nouns," he explains. A semantic databaseincludes object-types, which are like nouns, relationship-types that function as verbs,characteristic-types, which are like dependent clauses and property-types, which are likeadjectives and adverbs.

Elliot King is an Associate Professor of Communications at Loyola College
in Maryland. He can be reached at (410) 356-3943, or by e-mail ateking@loyolanet.campuscwix.net.