From niche market to mainstream?; object proponent: OODBMS superior to relational - object-oriented database management systems

From Niche Market to Mainstream?

The growing acceptance of the relational model as the fundamental platform for distributed database applications is a given. New and enhanced products to support distributed access and storage are being introduced at a rapid rate.

However, a small but vocal minority maintains that the relational model is flawed in that it is incapable of supporting the constructs required to store complex data stores such as voice, voice-annotated text, engineering drawings and graphics [See "Objects Are Taking Shape in Flat Relational World," Software Magazine, June, p. 32]. Their answer is object-oriented database management systems (OODBMS).

Object Design, Burlington, Mass., will ship the first version of its OODBMS, Object Store, in the fourth quarter, after beta testing in leading office information companies such as Microsoft, Lotus and NEC.

Tom Atwood, Object Design's founder and chairman, commented that less than 5% of corporate data today is automated because, he believes, less than 5% is record-oriented. And record-oriented data is what relational databases are designed to handle. So, he reverses the view that OODBMS will fill a niche market, stating that, "Relational can be considered the niche market."

Atwood notes that typical PC applications do not use relational technology because it is not fast enough. Relational technology is optimized to handle complex queries and searches through database tables, while Object Store is optimized for very fast traversal of specific applications' data structures once they have been moved into the computer's memory.

Atwood maintains that the object model is more capable than the relational model.

In terms of referential integrity, the object model is simpler to use, he said. The relational model is constrained by its dependence on only one modeling primitive, the tuple or record. However, there are at least two distinct semantic constructs: relations and objects. This has forced the addition of numerous rules to make some tuples behave like objects and others like relationships.

For example, you can say a person owns a car. There are two objects, the person and the car, "owns" is the relationship between them. If you delete the relationship, neither the person nor the car goes away; the person still owns the car. Thus, there is a semantic difference between the way relationships behave and the way objects behave.

Said Atwood: "A relationship is ontologically dependent on the prior existence of the two objects it relates." In the object modle, the relationship is designed into the definition of the object, thus integrity is an intrinsic part of the model.

Atwood also said that the query optimizer in Object Store is more sophisticated than those supplied with relational DBMSs. One reason for this is that their technology is 10 years older. Another reason is that the object model includes the operations as well as the objects, which enables it to, in a sense, anticipate the data required from the server and have it ready when the workstation requests it. Also, it makes more sense to move arbitrary pieces of programs to the processor where it can execute most efficiently, given some understanding of where the data is located.

The Object Store model moves data to the site of the execution of the process. But when it makes more sense to run the process somewhere else, Object Store uses Remote Procedure Calls (RPC) to relocate the SQL data manipulation sublanguage and arbitrary pieces of the program which include the data references.

Factors that will propel OODBMSs out of the niche market and into the mainstream include a 3:1 decrease in the costs to develop applications initially, and perhaps a 10:1 decrease in the life--cycle maintenance costs as libraries of reusable object types are built up, according to Atwood.

Time to market is a strategic issue for many companies. Having a library that already includes the object types that the company deals with means that only new objects required by the specific application have to be added to get a new product out the door.

He noted that there are three characteristic reactions to object-oriented technology.

First, people who have used the relational model for a long tie understand how much easier the object model is to work with. Second, people who are just starting to go relational are the most contentious. Third, those with a lot of big systems that are still IMS- and network-based find that the object model looks a lot like the network model. They find it intriguing to just skip a generation and go directly to object.

According to Atwood, the role of the mainframe will be relegated to that of the channel controller to the disk farm. And the traditional MIS organization will have to be drgged into this new object world "kicking and screaming."

COPYRIGHT 1990 Wiesner Publications, Inc.
COPYRIGHT 2004 Gale Group