UML: Ready for prime time? -- ADTmag

UML: Ready for prime time?

By Deborah Melewski
July 13, 2001

Just 18 months after the first iteration of the Unified Modeling Language (UML) object-oriented (OO) methodology first came to light, the specification has gained widespread support of the software development community and is expected to be declared an Object Management Group (OMG) standard. That effort has cleared a hurdle still facing other potential object standards -- strong backing from Microsoft Corp.

Nonetheless, UML has been criticized for its complexity, its lack of process, its bias toward the authors' methods and the difficulties facing tool vendors in providing full support for UML. In addition, many observers note that few software developers utilize any formal method for analysis and design, a key reason for the lack of long-term success by early CASE vendors over the past decade.

Work on the UML began in late 1994 when leading methodologists Grady Booch and Jim Rumbaugh joined forces at Rational Software Corp., Santa Clara, Calif., to create a single object-oriented methodology. By 1996, Ivar Jacobson, creator of the Jacobson object-oriented methodology, joined in the effort to create a unified method. The union of three leading OO methodologies prompted quick support from vendors of development software and other methodologists. For example, the latest version 1.1 of UML, now in the hands of the OMG, Framingham, Mass., results from a collaboration of large and small software suppliers and consultants, including Hewlett-Packard Co., IBM, Icon Computing, i-Logix, IntelliCorp, MCI Systemhouse, Microsoft Corp., ObjecTime Ltd., Oracle Corp., Ptech, Platinum Technology, Rational Software Corp., Sterling Software Inc., Taskon and Unisys Corp..

The UML is a visual modeling language comprised of a standard notation, semantics and a metamodel. The proposed standard defines the core concepts for object analysis and design (OA&D), and provides a standard set of diagrams for communicating those concepts. Backers of UML maintain that the convergence of technologies can signal an end to long-running methodology wars as well as provide model reuse and multivendor tool interoperability.

While most software tool providers and early

Rich Soley,
OMG chairman & chief executive

UML implementors say they are satisfied with the standardization effort so far, there remains some concern that the resultant methodology will not live up to its hype. However, backers say that in an industry that has seen promises of silver bullet solutions fail repeatedly, there is bound to be skepticism. Some observers are expressing concern that the end result may be of overwhelming proportions -- a language incorporating everything but the kitchen sink. Others are disheartened that the OMG halted the addition of what specific suppliers see as critical technology in an effort to speed the creation of the first rendition of a UML standard.

One critic of the unified language is object analysis and design tools supplier Select Software Tools, Irvine, Calif. The company is a long-time purveyor of the Rumbaugh OMT method and Jacobson's Objectory use cases, and recently aligned with Softlab, Atlanta, to incorporate the Softlab repository into the Select toolset. Stuart Frost, president and chief executive of Select, agrees that in notational terms, the UML is a natural step forward, but also expresses concern over the complexity of the results of the Booch-Rumbaugh-Jacobson collaboration. "When Booch and Rumbaugh first got together, they were able to keep the initial concepts of the unification," he said. "Unfortunately, too many others got involved, and in my mind, Booch and Rumbaugh lost some control and had to tear down a lot of their initial work." Frost likens the result of the collaboration to "too many cooks spoiling the broth."

"We recognize that you have to have standards -- that this is a necessary evil," said Frost. "But I feel that with the UML we've lost some of the elegance and simplicity of the initial plan and that the end result is clumsy, due largely to a lack of consensus."

Even so, Frost concedes that the UML has quickly become a de facto industry standard. "Everyone in the industry wants to use it," he said, while noting that the method remains a very small piece of the overall software development process. "Our more pragmatic, mainstream customers want to solve problems," Frost said. "They're not just looking at how the notation works. What they need is to wrap up the UML with a full methodology."

Frost is also disappointed with the UML's lack of component modeling. "Component modeling is a very critical piece for us, and unfortunately, components are an extension in the UML," he said. "Obviously, Select would like to see components at the heart of it."

Despite the criticism, Frost said that Select is committed to supporting UML, though it has yet to implement much of the latest version of the process into its product suite. "At the end of the day, it all comes down to what the customer needs," said Frost. "People do need standardization. But they don't mind if something is not 100% UML compliant. It's not like Java, where you have to have 100% compliance. The UML is designed so that every vendor can have different levels of compliance, based on what is needed."

A yeoman's effort

It's not possible to please all developers all of the time, a fact that is especially true in software standardization efforts. Mary Loomis, co-chair of the OMG OA&D Task Force and director of Palo Alto, Calif.-based Hewlett-Packard Co.'s Software Technology Labs, said that the OMG and Rational indeed faced a huge challenge in their effort to convince a large group of companies to agree on the core UML technology.

The OA&D Task Force was formed in 1995 to recommend technology built to improve interoperability between analysis and design tools and to ensure that models and designs could be

Mary Loomis,
co-chair of the OMG OA&D Task Force and director of HP's Software Technology Labs

moved between multivendor tools. The timing of the group's formation seemed right, with enough technology convergence to let developers find agreement on basic semantics and notation. The task force first decided to utilize a core technology comprised of pieces of a UML that could gain general support among the OMG membership. Parts of the UML that garnered the most criticism were not included in the submission of the task force, Loomis said. For example, the process used to develop models and specifically, the set of questions asked to develop a model, is not standardized.

"It was a bit scattered in the beginning. Everyone was using different terminology and notations," Loomis said. "We had to first get through the hurdle of standardizing on the notation itself, and then we wanted a fairly detailed meta model. This all had to be very precise and meant getting into some fairly detailed technical information."

After what Loomis describes as a yeoman's effort by a lot of people, and a lot of give and take, a final specification was approved. "The best minds in the business have spent a tremendous amount of energy on this," she said. "We feel now that we have good convergence and a good specification."

OMG acceptance

Once the method is approved as an OMG standard, Rational will turn over ownership of the UML trademark to the OMG. At that point, an OMG revision task force will be formed to handle omissions, problems, error fixes and overall maintenance. The OMG said the standard will include an extension mechanism that allows users to extend UML to meet proprietary needs.

The OMG's involvement in the proposed UML standard has come a long way since the group's first venture into the realm of methodologies in the early 1990s. "We did a study on object-oriented analysis and design tools about five years ago," said Richard Soley, OMG chairman and chief executive. At that time, "members looked at existing tools and published a report that compared methodologies. We took a lot of flak for that study. Users wanted to know what the differences were, but there was a cry that the OMG shouldn't get into standardization in that area. So we backed away."

About two years ago, however, Ivar Jacobson, creator of the Objectory method and use cases, pushed for a new standardization effort. OMG had the open process that facilitated this, and a large group of methodologists were brought together. Several submissions were received in response to an OMG request for proposals. Eventually the proposals were merged into a single submission.

The likelihood that a standard will be approved by year end 1997 "looks very good," said Soley. "We expect to see tremendous demand from tool vendors to support the UML as well as demand to move from existing models to the UML."

Soley warns, however, that the process is not all pink and rosy. "The promise of standardization is strong, but there will be a lot of work for vendors. We have a wide variety of companies involved to ensure this meets the needs of the customers. The main benefit will be a standard. At the core is a standardized meta-methodology, with standardized notation and a standardized API for tracking a methodology."

Early endorsers

Many companies have been supporting and working with UML for some time. In addition to the co-submitters of the UML proposal, many suppliers signed on as UML endorsers. Not

Dawn Stringfield,
Insight Technology Group chief consultant
and training director

surprisingly, many vendors are claiming to be the first in their product category to support UML in a product. But as Soley predicted, implementing UML is not always easy for vendors. Software suppliers are actually continuing to evaluate UML and are still deciding how much of version 1.1 will be incorporated into specific products.

One early endorser is Cayenne Software Inc., Burlington, Mass. Cayenne supports UML version 1.1 in its ObjectTeam 7.0 product, which was shipped to beta test sites in September. "It's virtually impossible for any vendor to implement 100% of the UML 1.1 spec," conceded Michael Jannery, senior director of marketing at Cayenne. "There are, however, five major diagrams in 1.1 that are clearly defined. That is what we can support."

The five diagrams, designed to support static and dynamic object models, include Use Case, Collaboration, Class, Sequence and State diagrams. According to Jannery, Cayenne has determined these five domains to be most important to users. He expects that the remaining oblique areas of the specification will be fleshed out over time.

Platinum Technology, Oakbrook Terrace, Ill., a cosponsor of the OMG proposal, has also announced product support for UML 1.1. Platinum had first submitted a rival proposal, but quickly retreated in the face of overwhelming support of UML. "As the OMG became more involved, it appeared that the responses were all somewhat similar enough that all of the submitting companies could work together," explained Dilhar DeSilva, Tools Integration Manager at Platinum and a member of the UML semantic task force.

Though the first UML standard will not incorporate many features favored by Platinum, DeSilva is convinced that the process will allow for changes later on. "We wanted to include other methods, like CDIF, to have a standard for moving data between different tools," said DeSilva. "We had to put a few things like that on hold, but once the UML is accepted, Platinum and other companies will be able to take on tasks that were dropped because of time constraints."

Platinum's Paradigm Plus analysis and design tool has supported UML since its release from Rational. Support for UML 1.1 will be included in the product upon approval of the standard, DeSilva said. Platinum is helping customers incorporate business process modeling into UML and offers a learning edition of Paradigm Plus that fully supports the UML. "Platinum saw the pain many of our users felt in having to decide on one methodology from among at least 20 methodologies out there," noted DeSilva. "We saw the need in the industry for people to come together and work from one metamodel."

For example, Cerner Corp., Kansas City, Mo., faced the methodology dilemma in 1993 when the company decided to move its DEC-based Cobol and C platforms to a client/server architecture. After reviewing a plethora of methods, the company chose the Rumbaugh OMT method. "There weren't a lot of tools out there at the time," said Doug Darbyshire, standards analyst at Cerner. "We eventually decided on Paradigm Plus [from Platinum technology], and when that product adopted Jacobson use cases, we did too."

Cerner develops software for hospitals and large health care organizations. The company's solution, which includes internally built middleware and SQL servers, automates health computer networks.

The Cerner offerings must adhere to FDA regulations that require extensive design controls, so the company must confirm that all required views were completed and that all major steps were taken. "We really need a full methodology, with analysis, design and full traceability of requirements to code," Darbyshire said. "We create state diagrams to show all the operations that we need to satisfy FDA regulations. The UML is not step-by-step. The intent is that you start with an immature model and then use the tools to make sure you have a mature object model. You must set a process in place to ensure that you reach that stage."

"The UML also gives us a platform for publishing our designs on the Web," Darbyshire said. "I can write the best reusable objects, but if I don't publish them it's of no use. The UML allows for publishing and communication between people."

Cerner has also overcome some basic design hurdles, such as when different people describe the same process in a different way. "Using Paradigm Plus with the UML allows us to automatically draw diagrams, and the repository allows for commonality," Darbyshire said.

Consulting/training needs

As with any new standard, UML brings with it a need for training and shared expertise. The good news is for users previously involved with the Rumbaugh OMT, Booch and Jacobson methods; the conversion to UML is somewhat easier than moving from other methods.

"We don't expect the average modeler to have to go read the OMG specification," said Loomis of the OMG OA&G Task Force. "People will be looking for help, and I expect to see a lot of books and training becoming available."

At least one consulting and training firm that specializes in object technology is already facing significant demand for UML services. Said Dawn Stringfield, chief consultant and training director at Insight Technology Group, St. Louis: "We have many clients using the Jacobson and Rumbaugh approach, and many of these clients want to move to UML primarily because they know it will become a standard soon and that most tools will incorporate it." Insight provides consulting, training and Insight Advantage, an application development process that can be used with the standardized notation of the UML.

Insight customers have been applying early versions of the UML on everything from order entry systems and billing systems to building controllers on manufacturing floors, Stringfield said.

In addition, many Insight clients in the financial services industry, which have traditionally depended on mainframe-based systems, are looking to utilize object technology in next generation systems. "We're suggesting UML because application developers want to have a standardized approach to go to full object-oriented technology," Stringfield said, noting that not only is a standards process nearly completed, but perhaps more important, it has the backing of "big players like Microsoft adopting it, and that really helps with acceptance."

According to Stringfield, the loudest concerns about UML come from clients that are veteran users of object technology, which want to be assured that past investments will be supported by the methodology. She said those concerns are easily allayed because "most tools have transition plans to support this."

"Still, there is such a smorgasbord of notations out there and people need to know how to capture and use them," Stringfield said. "There will be some new diagramming aspects, and some graphical changes required." Users moving from Booch to UML, for example, will find that core concepts stay quite consistent. The transition from other methods may be more difficult, she said. Stringfield complimented the progress of the UML, which is already providing users with benefits from standardization in terms of consistency, presentation and effective use of training dollars. "The UML brings a robust notation, especially in handling requirements design," she said.

Insight is aiding Hilco Technologies Inc., St. Louis, in implementing UML to build a new object-oriented, network-based manufacturing software system. Hilco previously used the waterfall method of software development and has long used the Yourdon method. "During the last two years, we've undergone a major cultural change in the way we do software development," said Bill Richards, product manager at Hilco. Because the company is using the UML only on its new projects, converging older methodologies or notations has not yet been an issue.

When Hilco decided to embark on object-oriented programming efforts, developers quickly determined that a structured method for building objects was essential. The company quickly selected UML in its software development and systems integration businesses and contracted with Insight for training. While the training took some investment, Richards noted that his developers are already in tune with the technologies.

"It's always a concern that a methodology can become overwhelming," said Richards. "We've found the UML is a really precise and user-friendly way of presenting use cases. It's also an easy way for the product development side to document user needs."

From the systems integration side, Hilco has found the UML invaluable in requirements design, and in getting customers to agree on requirements. "Beyond the use cases, we've found the UML is also good in communication of the architectural pieces. Because we're a big company it's easy to lose communications, but we've have been able to better maintain them with the UML."

Another early implementor of the UML is Southwestern Bell, St. Louis. Bob Gulledge, manager of middleware development, said he started working with UML at the end of 1996. "The UML is our modeling tool of choice and we're using it for the next release of our middleware product," Gulledge said. While he dabbled in other methodologies and modeling languages, Gulledge waited for the UML, whose release was on the horizon at the start of the project. "We waited for a working UML basically so that we
wouldn't have to switch later. If you use an older modeling language, you may also get stuck with older disciplines."

The main appeal of UML for Gulledge and his group was the unification of the methodologies of the so-called "three amigos," an industry nickname given to Booch, Rumbaugh and Jacobson. "It looked obvious that this would become the modeling language of choice. It became a de facto standard before it was even recognized," said Gulledge.

Members of Gulledge's Southwestern Bell unit sat in on early OMG discussions on UML, liked what they heard and bought the first available copies of the methodology. "We got real serious about it," Gulledge said. "Then we downloaded Rational's Rose product and started experimenting. We learned that Rational put a lot of work into the UML even before it was submitted to the OMG. The beta was somewhat buggy, but Rational was able to help us along."

While Gulledge is mostly satisfied with the UML, he noted that there are a few areas that still need improvement. "We're using the UML now for application modeling, but I'm also interested in learning business level modeling. I've found a few UML books out, though not enough yet, and there's just not a lot out there on the business modeling aspect."

While the UML does support business modeling, Gulledge has found documentation lacking. "Overall, the UML is quite powerful, but there are a lot of pieces of it that are undocumented," he explained. "You can go to the UML spec, but that's pretty tough to work with." There is also power in the UML that Gulledge said his unit isn't prepared to exploit. "There are a lot of good things talked about in the spec, but you really need to see these things applied to know how to best use them."

Gulledge said the costs of implementing UML can become an issue for corporations. Implementation requires substantial costly training, costs that Gulledge said could be alleviated with more local training firms and a proliferation of UML training in universities. The cost of Rational Rose was an issue, though Gulledge said, in the end, it was also critical to the project. The company is using the C++ and Java versions of Rose, which Gulledge noted is running them about $3,600 per copy. Despite early misgivings about the price, Gulledge said the cost was necessary to correctly implement UML. "While we decided to purchase Rose because we liked the tool and its use of the UML, there were other benefits that came into play including the tool's use of Java, and the Rational toolsets for requirements management. The less expensive products out there don't seem to have the full capabilities of Rational Rose," he said.

MCI Systemhouse, Atlanta, contributed to the UML specification and uses the language internally as a systems integrator. Cris Kobryn, Chief Scientist at MCI Systemhouse and chairman of the UML semantic task force, said: "We [designed] the metamodel for describing the semantics of the UML. This metamodel had to be very precise in defining what an object model means."

MCI Systemhouse's Knowledge Repository and Best Practices for Systems Integration CD-ROM now incorporate the UML, Kobryn said. "We've worked with many methodologies, and we recognized that there were complementary areas of leading methods that could be converged," Kobryn said. "We were combining the OMT and Booch methods, even prior to the three amigos, because we were working on solutions to the methods problem. We actively anticipated a UML and started working on a unified method several years ago."

From our perspective, the concept of the UML was a wise decision and consistent with our approach to a unified method," Kobryn said. For about a year now, MCI Systemhouse has used UML for the knowledge repository and for development projects. "It's a big win, because one spends a lot of time in projects just learning a common methodology. We've gained significant economies by agreeing on a common terminology."

Kobryn said the creators of the final UML specification have made a concerted effort to keep the language simple enough to be understood by average object-oriented programmers. "It can be a simple general purpose language, yet it is also scalable with advanced concepts for power modelers. In particular, there are some business process modeling extensions, and we paid attention to workflow issues as well."

The fact that so many competitors in the same market have signed up as partners is confirmation of the merit of the specification to Kobryn. "The fact that this is the first OMG effort that Microsoft has partnered with is also a show of the importance of the UML spec."

After considering several object modeling tools, MCI Systemhouse standardized on Rational Rose and is working closely with Rational.

According to Kobryn, there are still developers at MCI Systemhouse who do not spend much time modeling. Some may hack a system, for example, and begin programming without paying attention to analysis and design work. "I expect that the UML will help a lot in that area, and that further up the scale there will be an incremental process for more advanced constructs. Ideally, people should pick up more powerful features of the language as they go along."

Business process modeling

IntelliCorp, Mountain View, Calif., longtime proponent of the Martin-Odell object methodology, is now a backer of UML. Further exemplifying the unity involved in bringing about this technology convergence, IntelliCorp's lead methodologist, James Odell, co-author of Martin-Odell with James Martin, is co-chair of the OMG OA&D Task Force.

A proponent of business process modeling, Odell worked to ensure its incorporation into the UML in terms of diagramming techniques. Conrad Bock, methodologist at IntelliCorp, said there is enough process modeling in the final specification to let developers start working on the procedure. "We will have a revision task force after adoption of the spec, so we hope" to extend the capability.

IntelliCorp is studying the adoption of the UML, but has not as yet incorporated the specification into its own modeling product. Bock noted that there are two different worlds: business process modeling and object-oriented modeling. "It's hard to bring them together in the real world," he said, adding that IntelliCorp is evaluating the UML, and examining its strengths in business modeling. "Eventually, these two worlds had to come together."

IntelliCorp is mostly interested in the UML for its common set of concepts. "Even though the UML submission is still being evaluated at the company, we do believe it is exactly the right thing for the industry to be doing," confirmed Bock. "It will, however, take more than a year and a half to bring the two worlds together. I'm not displeased at all with the pace or the prospects of the UML. We're seeing a normal ongoing synergy, and there has to be a bridge between these two concepts and communities."

Aonix, San Francisco, already sells a UML version of its repository Software through Pictures (StP) analysis and design tool for complex operations. "For us, a natural evolution of our tool is to adapt to the latest and greatest methodology," said Larry Vernec, Aonix director of product marketing. The company offers StP for UML, along with a number of training courses on UML itself and on converting from OMT to UML.

Vernec agrees that the UML 1.1 specification may seem somewhat impractical because it is so broad. "It seems that the way the co-submitters solved disputes was to throw everything into the spec," he said. "The challenge for people will be to pull the particular subset that makes the most sense from them and their particular project. Each tool vendor has to decide how much of the UML to implement. For large projects, you need to implement all of the UML. Serious vendors will implement all of the UML right away. Other vendors may be trying to appeal more to the I/S market, working with smaller projects."

Aonix, like most vendors, is facing strong demand for UML support. Vernec, however, noted that many of these customers are not ready to use UML immediately so it may still be too early to predict how the standard will be used over the long term. "We've all heard the positive side of the UML," he said. "The concern is really the scope. Users have to learn to navigate the UML, and with the proper consulting and training, they can do that."

Room for improvement

Work will obviously be ongoing on the UML specifications. Extensions will be carried forward, and work in new areas is progressing under the OMG. For example, ObjecTime Ltd., Kanata, Ontario, Canada, and IBM jointly submitted a proposal to ensure definition of special-purpose extensions to the UML.

Of particular interest to ObjecTime is the UML-RT, a realtime extension to UML based on the company's Realtime Object-Oriented Modeling (ROOM) methodology. Garth Gullekson, methodologist and founder of ObjecTime and program director of the company's UML activities, finds no discontinuity in aligning with the UML. While the ROOM method is proprietary, the UML is expected to help the company's customers deal with both realtime and general purpose modeling problems. "The knowledge our customers have in ROOM will be transferable to their work under UML," he said.

A component modeling extension is already being pushed by several suppliers, including Select Software, for future versions of the UML specifications. "The fact is there are a lot of components out there and Select has some good technology in that area," said Tony Baer, IT [information technology] Consultant at Demand Strategies, Bedford, Mass. "But the UML is a language, and it may be difficult to incorporate components at this point because there has to be a way to represent content."

In addition, Baer noted, the OMG faces a problem because components are hard to define. "Even if you can put a notation around what it is, you may still have a problem in defining the content," Baer explained. "For example, if you put a couple of
JavaBean components together, what's to keep the components from overlapping? Any multivendor standards process is always going to be a compromise," Baer said. "Interesting in the UML process is that former rivals found themselves working together to co-sponsor the same proposal."

Baer calls the UML a rare case of farsightedness in the industry. "Customers had a very influential voice and demanded that companies get their modeling act together. Obviously there was a growing commercial use for the UML, and the timing was right to get people together," he said.

It may be too soon yet to see the full impact of the UML or to recognize potential problem areas. Looking at the overall picture, however, it would appear that great effort went into the specification and that many companies have taken an early lead in implementation of the soon-to-be standard.

"While the UML may be considered a compromise, it's important to look at it in the grand scheme of things," said Baer. "It's obviously better to have 80% of a solution than no solution at all."

While most software tool providers and early UML implementors say they are satisfied with the standardization effort so far, there remains some concern that the resultant methodology will not live up to its hype. However, backers say that in an industry that has seen promises of silver bullet solutions fail repeatedly, there is bound to be skepticism. Some observers are expressing concern that the end result may be of overwhelming proportions -- a language incorporating everything but the kitchen sink. Others are disheartened that the OMG halted the addition of what specific suppliers see as critical technology in an effort to speed the creation of the first rendition of a UML standard.

A yeoman's effort

The OA&D Task Force was formed in 1995 to recommend technology built to improve interoperability between analysis and design tools and to ensure that models and designs could be moved between multivendor tools. The timing of the group's formation seemed right, with enough technology convergence to let developers find agreement on basic semantics and notation. The task force first decided to utilize a core technology comprised of pieces of a UML that could gain general support among the OMG membership. Parts of the UML that garnered the most criticism were not included in the submission of the task force, Loomis said. For example, the process used to develop models and specifically, the set of questions asked to develop a model, is not standardized.

OMG acceptance

Early endorsers

Many companies have been supporting and working with UML for some time. In addition to the co-submitters of the UML proposal, many suppliers signed on as UML endorsers. Not surprisingly, many vendors are claiming to be the first in their product category to support UML in a product. But as Soley predicted, implementing UML is not always easy for vendors. Software suppliers are actually continuing to evaluate UML and are still deciding how much of version 1.1 will be incorporated into specific products.

Cerner develops software for hospitals and large health care organizations. The company's solution, which includes internally built middleware and SQL servers, automates health computer networks.

Consulting/training needs

Insight customers have been applying early versions of the UML on everything from order entry systems and billing systems to building controllers on manufacturing floors, Stringfield said.

After considering several object modeling tools, MCI Systemhouse standardized on Rational Rose and is working closely with Rational.

Business process modeling

Room for improvement

"While the UML may be considered a compromise, it's important to look at it in the grand scheme of things," said Baer. "It's obviously better to have 80% of a solution than no solution at all."