A Rational Approach to Objects

As this article went to press, the "three amigos," as they've dubbed themselves, were preparing a proposal to the Object Management Group (OMG), Framingham, Mass., to certify their combined work, called the Unified Modeling Language (UML). The proposal is in response to an OMG request for technology that calls for a standard language and meta model that will allow different modeling tools to interoperate. Oracle, Hewlett-Packard and Texas Instruments Software have endorsed Rational's OMG submission. The kicker? So has OMG's nemesis, Microsoft.

To cynics, there may appear to be something wrong with this picture -- the object tools market agreeing on a standard that establishes the playing field on which everybody competes? Imagine where Unix would be today if IBM, Digital and Hewlett-Packard had formed such a consensus with Sun Microsystems a decade ago.

Certainly, the battle isn't over, but a dramatic consolidation has already occurred. "There were so many people writing about object-oriented analysis methods that it became too much," says Jim Odell, a consultant based in Ann Arbor, Mich. Odell himself has written an object analysis and design method.

In the end, the market drove the consensus, says Jim Smith, COO and executive vice president of sales and marketing at Select Software Tools Inc. in Santa Ana, Calif., one of Rational's rivals. "When you combine three of the top methodologies whose methods have reached over 50% penetration, that gets attention," he says.

Step one of the consolidation occurred in 1994 when Rumbaugh left General Electric's Schenectady, N.Y., research center to work with Booch at Rational. "Before we came together, we were in a kind of rivalry. We tended to exaggerate the differences for our own benefits," says Rumbaugh.

In fact, the Booch and Rumbaugh methods were pretty complementary, according to Andre Lecl?rc, president of Technology Development Associates, Morganville, N.J., an object technology development firm. Booch provided a way of representing large, complex systems, while Rumbaugh provided a simpler notation and a way of relating object models to the database world.

Meanwhile, Jacobson, who had formed his own company, Objectory AB in Stockholm, Sweden, developed a form of process analysis -- "Use Cases" -- that took a new approach to analyzing systems. This approach doesn't analyze systems by looking at the way databases or applications interact, but as business processes performed by real people. "It started by looking at the system from the outside, not the internal details of how you design it," says Jacobson. In 1995, Rational bought Objectory, setting the stage for the unification.

Ironically, Rational's major rivals have also embraced the Booch, Rumbaugh and Jacobson methods, in some cases faster than Rational itself. Even tools that aren't based on object-oriented development, such as Texas Instruments Software's Composer, will support UML as part of a component strategy. Today, Select OMT from Select Software, Paradigm Plus from Platinum Technology Inc., Oakbrook Terrace, Ill., and ObjectTeam from Cayenne Software Inc., Burlington, Mass., support various draft versions of the UML, while Rational won't extend support until version 4.0 of Rational Rose, slated for release late this year. Furthermore, the company won't fully integrate the Objectory Use Case capabilities until it releases one or two more versions following Rose 4.0. According to Booch, Rational didn't feel that the UML was stable enough to revamp Rose until now.

Rational's lag in supporting UML is also a matter of expediency, adds Michael Merriman, an analyst at Gartner Group, Stamford, Conn. Because Rose was written directly around the methodology, it takes more effort to update it than it does for competing products such as Select OMT, Paradigm Plus and ObjectTeam, which have a more loosely coupled architecture. Nonetheless, adds Liz Barnett, an object technology analyst at Giga Information Group, Cambridge, Mass., Rational's delay in supporting UML "is a real exposure."

And, while UML is fast becoming the de facto standard object modeling language, peace hasn't broken out everywhere yet.

Platinum has formed the Open Object Alliance to submit a rival proposal to OMG that will add some notations to UML related to interoperability, as well as a different meta model. The likely outcome is that OMG will support a combination of the submitted proposals, according to IDC's Garone. Significantly, Platinum has committed to supporting OMG's decision, but Rational thus far has not.

The convergence to a common modeling language, such as the UML, reflects the fact that language alone is not the major contribution of object tools. Instead, it is factors like ease of use; support of popular client/server development tools, databases, and platforms; and features like reverse engineering and integration with data modeling tools that will matter in the long run.

Does any of this standards hoopla really matter? Objects have been cursed with the "next-generation" label since the '80s. The early view of objects was that they required hard-to-learn, poorly supported languages like Ada, C++ or Smalltalk, and that they relied on obscure, object-oriented databases that neither scaled well nor adequately supported business transaction processing applications.

However, fast-forward to the present, and objects are being called upon -- not to replace client/server tools, databases or even legacy applications, but to help developers deploy and integrate them more consistently. Some companies, like Xerox, are relying heavily on object-based, client/server migration. As objects gain use, modeling language convergence will take on new importance.

A major use of object modeling is to help developers scale client/server applications. For instance, a relatively simple stock quotation system that collects data in realtime and displays it on a spreadsheet could be finished with a rapid application development tool in a matter of weeks. However, by the time the system is extended to the fourth or fifth business unit and transaction volumes go up, developers are faced with the issue of how to distribute the application to maintain performance. Figuring out how to do this without modeling amounts to guesswork, says Steve Bailey, principal consultant at Tier Corp., a Walnut Creek, Calif.-based client/server consultancy. "People often wind up with duplicate databases and data entry forms, and they find it hard to identify all the sources of data and business logic," he says.

Ideally, OO analysis and design tools should provide a blueprint of application architecture, including identifying the business objects and what functions, data and data sources they involve. In distributed systems, the tools should also reflect where different business objects or application functions should reside.

According to Gartner's Merriman, object design tools have not yet perfected the art of modeling the front end of a system or adequately managing application partitioning. Furthermore, although most object tools have capabilities for generating code for languages or tools such as C++, Java, FortZÿ, PowerBuilder or Visual Basic, these capabilities are limited, according to Tier's Bailey. He notes, for example, that Rose's PowerBuilder facility generates functions and objects, but does not produce Powerscript or Data Windows for the front end. That limitation may be acceptable to many developers, however, because they may not trust object tools to generate their code or database connectivity.

Many of these issues are likely to become academic. A new generation of 4GL-oriented object modeling tools, which provides more seamless integration with popular client/ server tools, is on the horizon. Meanwhile, a more radical approach to object modeling -- where the functions of object modeling tools are deconstructed to pop up as design or architecture wizards to popular low-end tools such as Visual Basic -- is under development. In fact, the stage has already been set by Visual Basic Controls (VBXes), low-cost application functions such as bare-bones spreadsheets or user interface features such as radio buttons or tabs. Visual Basic developers have used these low-cost components to assemble a wide variety of inexpensive business applications.

Meanwhile, object models may become yet another kind of component. Developers could invoke an architecture wizard that provides a standard template for distributing application functionality, or a wizard for defining a simple business object. In fact, these components may not necessarily be called objects, and the process may not even be called object analysis or design, says Paul Levy, Rational chairman and chief executive officer. Rational is firing the first shot, thanks to a recent cross-licensing agreement with Microsoft that repackages specific capabilities of Rational's Rose product with all of Microsoft's development tools, beginning with the upcoming release of Visual Basic 5.0, to be released in January. In turn, future releases of Rose will share core Microsoft technologies, including the repository developed with TI; a common front end, Microsoft Developers Studio; ActiveX; and Distributed Component Modeling (DCOM). As moves such as these by Rational and others succeed in furthering object technology acceptance, the artificial boundary separating object models and application development may well disappear.

Tony Baer is a freelance writer based in Bedford, Mass.

November 1996/Tony Baer

Software Magazine's Tony Baer recently interviewed Paul Levy, Rational's chairman and chief executive officer, and Michael Devlin, president, to get their views on the object analysis and design market. What follows are excerpts from that interview.

Software: What is the role of object modeling?

Devlin: Besides the traditional role of enabling high-quality designs, modeling technology has become a technology that lets you visualize the non-visual elements of your application, like the business logic. In the last five years, we have seen dramatic progress from tools like Visual Basic, which showed how you could visualize parts of your application. Object modeling lets you visualize other components of your technology beyond the front end.

Software: What are the roles of Grady Booch, Jim Rumbaugh and Ivar Jacobson at Rational?

Levy: Today, they are members of the R&D staff, working with customers, partners and developing new technology. The UML is only the most visible part of what they do. Two years ago, we might have emphasized that we had Booch and Rumbaugh to help us get into large accounts. But today, the technology is becoming more mainstream and less personalized. In our [1996] annual report, we didn't even mention them.

Software: How is the object modeling market changing? Will it still be known as object modeling?

Levy: There won't be a design tools market anymore. It will all be based on visualization. Rose will become an indispensable tool to every visual application developer. But they won't think of it as a design tool. They will see it as a tool that allows them to graphically manipulate and manage every application.

Devlin: We're not trying to teach every application developer the theory of objects. We're shifting to a new paradigm of component-based development, where components are assembled using architectures and patterns as a guide. There's been lots of work done on design patterns to address technical software engineering problems. Rational and others will provide repositories of those objects.

Software: Does this mean downloading architectural "widgets?"

Levy: We are repartitioning our functionality to match price points to the needs of different parts of the market, like the entry-level Visual Basic or Visual J++ developer. In the future, you'll see revolutionary improvement in the usability of Rose, so it looks like Word or Excel. The user may not even know that they're building object models.

Devlin: It won't necessarily be a standalone product that's sold at Egghead. It is likely that we'll make Rose available directly to every VB developer through repackaging or easy downloads via the Internet. It could mean downloading from the Internet things like $49 common design patterns, executable wizards for designing three-tier applications, or small tools for generating test cases. You could then add sets of components to these architecture templates to make real business applications.

Software: What are Rational's plans for penetrating the mass market?

Levy: We'll embrace subscription and simple upgrade models. We will very heavily leverage our Web site as a distribution and support channel. We envision working with Microsoft across a wide range of selling and marketing initiatives, with Rose and other products as indispensable supplements to VB, Visual J++, Visual C++. Over the next year, you'll see several announcements from us that involve tools that we might develop or acquire.

Software: The mass market strategy sharply diverges from your traditional, high-end client strategy involving products and services costing tens or hundreds of thousands of dollars. Will Rational continue to serve its high-end client base, and if so, how will you avoid losing focus?

Levy: Addressing both the mass market and enterprise projects will not necessarily pull the company in conflicting directions. We've learned over time that products have to be more self-sufficient. We also believe that large customers will issue blanket purchase orders for large-volume tool sales from our Web site.

Devlin: Product ideas often start with high-end projects, and then proliferate as commodities.

Software: Can Rational make money at lower price points?

Levy: When we added telesales, we moved deal volumes higher, but at smaller $5,000 to $40,000 price points. We already have a lower cost-per-revenue dollar than many smaller companies in this industry and therefore became even more profitable. We believe that as we turn the crank on newer channels, we'll be able to improve our cost structures further.

Software: Will Rational support whatever emerges from the Object Management Group process?

Levy: OMG will be particularly important for driving standards across the different Unix variants. But Microsoft is obviously more important for the Windows community. It owns the de facto standards there. Our aim is to work with OMG to ensure that OMG and the market-share leader are one and the same. We'll drive the de facto standard and work hard with the formal standard. As long as what OMG comes up with helps us establish a broad worldwide standard, we'll be onboard.