Software Architecture (NOSA´99), Jan Bosch (Ed.), August 12-13, 1999.
An Evolution-oriented Architecture
for Web Applications
Guntram Graef and Martin Gaedke
Telecooperation Office (TecO), University of Karlsruhe, Vincenz-Prießnitz Str. 1, 76131 Karlsruhe, Germany,
email: {graef, gaedke}@teco.edu
Abstract: The Web has become an efficient environment for application delivery. The
originally intended idea, as a distributed system for knowledge-interchange, has given way to organizations offering their products and services using the Web as a global point of sale. Although the arising possibilities look promising, the development process remains ad-hoc in real-life Web development. The understanding of Web application development mostly neglects architectural approaches, resulting in Web sites that fail in achieving typi-cal goals like evolvable and maintainable structures of the information space. Beyond that, as the architecture of a Web application matures, more and more knowledge about the do-main becomes embodied into code and therefore burdens do-maintenance and reuse of parts of the application. In this paper, we will propose an architecture and a framework using the notion of services as model entities for Web application development. The object-oriented WebComposition Markup Language, which is an application of the XML, will be pre-sented as basis for a generic evolvable framework for services. Finally, the results of its us-age will be described in detail by giving an example of a large-scale transnational Intranet, where the framework is in use.
Keywords: Web Engineering, XML, WebComposition, Framework, Service.
1 Introduction
The original purpose of the World Wide Web [B-L90] was merely presenting information. Many of today's Web applications have moved a long way from the basic hypermedia systems of the first days. They have grown into com-plex distributed applications. Nevertheless, the development process in real life remains rather ad-hoc. This has largely been recognized to be due to the gap between the underlying design models and the implementation model of the Web. Web developers suffered from this when they tried to apply disciplined hypertext devel-opment methodologies such as OOHDM [SRB96] and RMM [ISB95] or general models like UML [Lar98] because many design con-cepts where lost during the implementation. This problem has been addressed with systems
like JESSICA [BaS98], W3Objects [ICL96] and WebComposition [GWG97].
The WebComposition approach allows a de-veloper to work within a semantically richer component model and automatically map the results to the Web implementation model. The mapping can be to any target language and is independent of the Web-server platform used. The WebComposition model enables and en-courages reuse of code on the component level. In contrast to reusing Web resources, compositional reuse is not limited to a certain level of granularity imposed by the Web im-plementation model. Furthermore, it is possi-ble to reuse abstractions that have been defined based on concepts such as generalization and specialization.
A major problem that remains is how to effi-ciently reuse larger structures when building Web applications from components. Pressure to do so is especially strong in the Web envi-ronment due to much shorter development and maintenance cycles than in most common software application domains. Questions like the following must be considered: What ab-stractions exist above the component level that are suitable for reuse? In what kind of devel-opment artifacts can these abstractions be cap-tured? What does a mechanism look like that enables their reuse? In which way can the re-use process be optimized for productivity? How can it be implemented? What are the real benefits that can be achieved?
In the next section, we will describe how ser-vices can be used as higher-level abstractions and coherent building blocks for Web applica-tions. We will also discuss frameworks for building Web applications. In the third section, we will introduce our own approach. We sug-gest a Service-oriented Framework that is based on the WebComposition approach. Fur-ther, we introduce the concept of a Service Factory to enable a disciplined evolution of the application. We will describe our concrete implementation in section four and give an evaluation based on theoretical aspects, fol-lowed by an example of a commercial applica-tion in secapplica-tion five. In the last secapplica-tion, we will conclude with a short summary and outlook for further research.
2 Architectural abstractions for the
Web
At first, we introduce an abstraction for func-tionality provided by a set of components for one or more applications. Then we will discuss frameworks as an abstraction on the domain level of applications.
2.1 Services
One aspect that distinguishes Web-applications from most common desktop ap-plications is that there is no "application in the box" distribution model. What the user re-ceives and pays for is a set of services that might more or less loosely be integrated or
federated into an application such as described in [GaT99] or [Zhao97]. Well-known exam-ples are portal sites such as Netscape’s Netcen-ter [NNC99]. On the Web, this grouping can change quickly and the same services might appear in numerous applications. In this sense, the application might even be built by users according to their own preferences or it might be dynamically generated to address a specific target group [FrK97].
In this context, a service presents a sub-system that enables a user to perform a single task such as placing an order or retrieving a news summary. In a business environment, a service may represent a business procedure.
Services can be classified according to their functionality. There might be services for re-trieving information, services for ordering products or services for expressing opinion such as online surveys just to name a few ex-amples.
The evolution of service classes tends to be subject to other forces of change than both the application and lower level building blocks. The shape of the application is directly driven by the current needs of its users that are closely related to sociological forces such as fashion. Lower level elements, such as components, that serve as building blocks for services are more directly affected by technological change. A class of services on the other hand might be mainly affected by the evolution of general business behavior.
This orthogonality as well as the inner coher-ence of services and classes of services makes them powerful abstractions for building Web-applications. They appear to be suitable enti-ties for reuse.
2.2 Frameworks
Very flexible abstractions for reuse on the level of an application domain are frameworks. As stated in [BMA97] a framework is an inte-grated set of software components that can be reused to create applications. The construction of applications is possible either by extension of framework components [Jon92] or by cus-tomization through parameterization of
func-tional specifications that are used to automati-cally build applications from framework com-ponents [Pre94]. These two approaches to framework reuse are called "white-box" and "black-box" reuse [RoJ96].
There exist frameworks for the Web such as for shopping- or e-commerce. Common Web frameworks typically use a black-box core consisting of platform dependent components residing on the Web-server and a white-box part made up of Web-resources. This reflects the notion of [JoF88] that both modes of framework reuse are simultaneously present in most frameworks. The problem is that the white-box part here suffers from the limita-tions of the Web-implementation model as described in [GGS99]. Furthermore, a very common limitation in today’s frameworks for the Web is that code from the white-box part can’t be migrated into the black-box part of the framework because of disparities in the im-plementation models. Well known examples for this kind of frameworks are [Int99] or [LDM99].
A more advanced application development framework for the Web should offer an inte-grated implementation model for the black-box core as well as for new components developed with the framework. This is needed to facili-tate the migration of new components into the black-box during the evolution of the frame-work. An integrated implementation model is also an important prerequisite for enabling improved reuse and maintenance of domain components.
3 Service Development Framework
In this section, we will introduce a Framework for developing Web services that we built based on the WebComposition approach.
3.1 Basic Architecture
The WebComposition approach presents a generic intermediate model to bridge the gap between fine-grained design and the coarse-grained implementation of Web applications by automatically mapping components to Web resources. WebComposition allows the defini-tion of aggregadefini-tion and inheritance reladefini-tion-
relation-ships between components based on a proto-type-instance object model such as in SELF [UnS87].
Components are defined using the WebCom-position Markup Language (WCML). They are stored in a Component Store from where they can automatically be mapped to Web resources using our WCML compiler. It is possible to generate code in any target language. Due to the generic approach, the mechanism is com-pletely independent of the Web server plat-form. Service Definition Service Factory WCML Compiler Web Resources Component Store
Service ComponentsFramework Service Development Framework
Figure 1 - Architecture of Service Framework
All components are described in the same way, including those in the black-box part of the framework and components developed by the framework user. This means there is an inte-grated implementation model enabling easy migration of components from the white-box part of the framework into the black-box core, facilitating framework evolution.
Services can be built by instantiating frame-work components. A service consists of com-ponents for content, layout, navigation, user interaction and processing. During run-time, the code produced by these components is called from the surrounding framework as needed. This general framework behavior is
usually described as the Hollywood Principle "Don’t call us, we’ll call you." [Lar98].
Order Service Service Navi-gation Layout User Interaction Content Pro-cessing Order Content Order Pro-cessing 1 * 1 1 1 1 * * * * * 1 1 *
Figure 2 - Generic architecture of order services
3.2 Factory-oriented Evolution
A central part of our framework is the concept of a Service Factory, which is based on cre-ational design patterns as described in [GHJ94]. It is intended to facilitate the devel-opment of common types of services. The Ser-vice Factory enables the reuse of generic archi-tectures of classes of services.
As an example, figure 2 shows a simplified description of the generic architecture of order services. An order service component is based on the service component, which it extends. Order services contain their own variant of content and processing components.
To create a specific service a service definition must be supplied consisting of various infor-mation items. It must define how many com-ponents of each required type such as process-ing components the service should contain. It also describes attribute values for all the com-ponents belonging to the service. Thus, it de-fines all aspects of the service that distin-guishes it from other services of the same ser-vice class.
The Service Factory parses the service defini-tion and uses its knowledge about the generic architecture of the corresponding service class to automatically create the service.
The Service Factory itself consists of a Factory Control Function that dispatches the service definitions to the appropriate Factory Methods
(Figure 3). The Factory Methods implement the Factory Method creational design pattern. There is at least one Factory Method for each service class capturing its generic architecture. Several sets of Factory Methods can be sup-plied alternatively as described by the Abstract Factory design pattern.
F a c tor y C ont ro l F u n c ti on Order Service Factory Method Information Service Factory Method
…
Figure 3 - Service Factory Architecture
4 Implementation
The implementation of the service develop-ment framework as well as that of the underly-ing WebComposition system rely on the eX-tensible Markup Language (XML) [W3C99] and related tools.
XML is a subset of SGML [ISO86] and pre-sents a standard mechanism for defining markup languages. It provides basic syntax conventions that are familiar to a large number of users, especially in the Web environment. It enables formal syntax definitions that are used for automatic document checking. There are numerous tools available such as parsers and editors that reduce the development effort usu-ally associated with the introduction of a new language. The WCML is an XML application and enables the definition of components. This facilitated the development of the WCML compiler and other tools since they can use standard XML parser components for docu-ment parsing.
When implementing the WCML compiler and tools, such as a Component Repository we introduced in [GRG99] and a visual editor for service definitions, we explored two
alterna-tives. We successfully used the XJParse XML parser [Dat99] for developing tools in Java as well as a COM [MSC99] based environment. Both implementation strategies worked well to produce tools that interoperate seamlessly.
5 Evaluation
In this section, we will first discuss general aspects of the Service Development Frame-work before we will briefly describe its use in a large-scale e-commerce application at Hew-lett-Packard.
5.1 General reflection
Our approach offers several levels of reuse. In addition to individual components, larger structures, namely service class architectures and knowledge captured in the framework’s component architecture, can be reused.
We separate code needed for describing indi-vidual services, code common to classes of services and framework code, which is com-mon to all services. This helps to reduce the complexity of the code that has to be main-tained by avoiding some unnecessary redun-dancy. Furthermore, specialized programming and teamwork are encouraged due to the task-oriented closure of services. On the one hand, this can make the development of specialized tools, such as a visual editor for a special but very common type of service, feasible. Specialized tools in contrast to general tools only need to cover a limited problem domain and thus are less complex. On the other hand, specialized programming can lead to a flatter learning curve and lower cognitive load for programmers. This can reduce technical skill requirements enabling a stronger focus on do-main expertise. It also can lead to higher pro-ductivity, better quality and faster deployment. Another feature of the framework is automatic code generation by the Service Factory and the WCML compiler. In practice, this greatly in-creases productivity. For example, a service definition describing a service for ordering office chairs can be made up of about 50 tags. The code produced by the framework can eas-ily reach more than 3000 lines of Web code mostly consisting of complex scripting
state-ments. We want to stress that in contrast to many code generator tools for the Web this code never needs to be modified directly and can always be regenerated automatically
Reuse, automation and reduced complexity contribute to enhanced maintainability and extensibility of services, applications and the framework itself. Maintenance also benefits from the central location of code common to classes of services or all services defined within the framework. Changes of corporate identity, process logic or technical infrastruc-ture only need to be implemented once and needn’t be repeated in several locations or re-sources.
Migration of existing Web applications and services to the framework is supported by the fact that existing code can be wrapped in WCML components and needn’t be repro-grammed. It is also not necessary to migrate applications all at once. It can be done in an incremental approach because framework gen-erated Web resources seamlessly integrate on the Web server with resources of other origin. In our approach, we also gain a high degree of flexibility for migrating to new implementa-tion technologies or for concurrent support of different implementation models. This can be achieved by encapsulating implementation specific code in special WCML components that can be dynamically replaced when gener-ating code for different implementation plat-forms. The service development framework itself and WCML are entirely independent of the deployment platform.
5.2 Eurovictor II
The Service Development Framework has al-ready been used to fully implement a complete service-oriented e-commerce application. Eurovictor II has been developed at the Tele-cooperation Office at the University of Karlsruhe in a joint project with Hewlett-Packard (HP). The system has recently become productive at HP in Europe with more than 10000 regular users.
The main goal of Eurovictor II was to enhance usability, evolution, maintenance, creation and
management of Web-based services and fed-eration of different intranet services within the company. The system already offers several hundred services mainly in the areas hardware orders, software orders, telecommunication, virtual office and news services. Services have been developed in a decentralized manner by people with moderate technical skills and have automatically been integrated into the system by use of an installation-service.
Figure 4 Screenshot of a temporarily available service of the Eurovictor II system
The quick and easy way in which new services can be developed has not only improved pro-ductivity but has also given way to completely new applications not possible before. An inter-esting example of this has been a service that has been set up in less than one hour for the only purpose of enabling the ordering of pizzas by employees in one HP location during a sin-gle day to celebrate a special event at the com-pany. We consider this to be an example of how an organization and their members can quickly react upon a new situation in a flexible way.
6 Conclusions and future work
In this paper, we introduced a generic compo-nent based framework supporting the devel-opment of web services.
We have shown how services can be used as building blocks for applications. By means of a Service Factory, we have been able to demon-strate how the generic architecture of a class of services can be reused.
In an evaluation we pointed out the various advantages of our approach that lead to im-proved productivity, better quality and cost reduction during development, maintenance and evolution of web applications.
We could verify our concept building a large-scale e-commerce application at Hewlett Pack-ard in Europe.
In the near future, we plan to further improve tool support. Our research will include auto-matic mining and classification of code struc-tures on the web that can be used as low-level building blocks for services. We are also inter-ested in exploring the dynamic assembly of web applications based on user profiles.
Acknowledgements
The authors would like to thank U. Stegemüller of the Hewlett-Packard Corpora-tion, Germany for his cooperation and thought-ful feedback for some of the ideas developed in the approach presented here.
References
[BaS98] R.A. Barta, M.W. Schranz (1998).
JESSICA: an object-oriented hypermedia publishing processor. In
Computer Networks and ISDN Systems 30 (1998), Special Issue on the 7th Intl. World-Wide Web Conference, Brisbane, Australia, April 1998, pp. 239-249.
[B-L90] Tim Berners-Lee (1990). World
Wide Web: Proposal for a HyperText Project,
http://www.w3.org/Proposal.html [BMA97] D. Brugali, G. Menga, A. Aarsten
(1997): The Framework Life Span. In: Communications of the ACM, Vol. 40, No. 10, Oct. 1997, pp. 65-68
[Cle96] P. Clements (1996). A Survey of
Architecture Description Languages.
In Proceedings of 8th International Workshop on Software Specification and Design (IWSSD), Paderborn, Germany
[Dat99] Datachannel Inc. (1999). XJParser. http://xdev.datachannel.com/downlo ads/xjparser/
[FrK97] L. Frelechoux, T. Kamba (1997). An
architecture to support personalized Web applications. In: Computer
Networks and ISDN Systems 29, Special Issue on the 6th Intl. World-Wide Web Conference, Santa Clara, CA, USA.
[GaT99] M. Gaedke, K. Turowski (1999).
Generic Web-Based Federation of Business Application Systems for E-Commerce Applications, 2nd Intl.
Workshop on Engineering Federated Information Systems (EFIS’ 99), Kuehlungsborn, Germany
[GGS99] M. Gaedke, H.-W. Gellersen, A. Schmidt, U. Stegemüller, W. Kurr (1999). Object-oriented Web Engineering for Large-scale Web Service Management. In: R. H.
Sprague (Ed.) Proceedings of the 32nd Annual Hawaii International Conference On System Sciences, Maui, Hawaii, (CD-ROM).
[GHJ94] E. Gamma; R. Helm; R. Johnson; J. Vlissides (1994). Design Patterns:
Elements of Reusable Object-Oriented Software. Reading.
[GRG99] M. Gaedke, J. Rehse, G. Graef (1999). A Repository to facilitate
Reuse in Component-Based Web Engineering. International
Work-shop on Web Engineering at the 8th International World-Wide Web Conference (WWW8), Toronto, Canada. http://budhi.uow.edu.au/ web-engineering99/accepted_papers/ gaedke.html , June 1999 [GWG97]H.-W. Gellersen, R. Wicke, M. Gaedke (1997). WebCompostion: an
object-oriented support system for the Web engineering lifecycle. In:
Computer Networks and ISDN Systems 29, Special Issue on the 6th Intl. World-Wide Web Conference, Santa Clara, CA, USA, pp. 1429-1437.
[ICL96] D.B. Ingham, M.C. Little, S.J. Caughey, S.K. Shrivastava (1996).
W3Objects: Bringing Object-Oriented Technology To The Web,
The Web Journal, 1(1), Proceedings of the 4th International World Wide Web Conference, Boston, USA, pp. 89-105
[Int99] Intershop Communications Inc. (1999). Intershop 3.0
http://www.intershop.com
[ISB95] T. Isakowitz, E.A. Stohr, P. Balasubramaninan (1995). RMM: A
Methodology for Structured Hypermedia Design,
Communi-cations of the ACM, Vol. 38, No. 8, August 1995, pp. 34-44.
[ISO86] International Organization for Standardization (1986). Information
Processing - Text and Office Systems - Standard Generalized Markup Language (SGML). Ref. No. ISO
8879:1986.
[JoF88] R. Johnson, B. Foote (1988).
Designing reusable classes. Journal on Object Oriented Programming 1,
5, June 1988, pp. 22-35
[Jon92] R. Johnson (1992). Documenting
frameworks using patterns. In:
Proceedings of OOPSLA 92, Vancouver, Canada, pp. 63-76
[Lar98] C. Larman (1998). Applying UML and Patterns. Prentice Hall, pp.
455-486
[LDM99] IBM Inc. (1999). Lotus
Domino.Merchant.
[MSC99] Microsoft Corporation (1999).
Component Object Model (COM).
http://www.microsoft.com/com [NNC99] Netscape Communications Inc.
(1999). Netcenter.
http://www.netcenter.com
[Pre94] W. Pree (1994). Design Patterns for
Object-Oriented Software Develop-ment. Addison-Wesley
[RoJ96] D. Roberts, R. Johnson (1996).
Evolving Frameworks - A Pattern Language For Developing Object-Oriented Frameworks. Third
Con-ference on Pattern Languages of Programs (PLoP ’96), Monticello, Il-linois, USA
[SRB96] D. Schwabe, G. Rossi, S. Barbosa (1996). Systematic Hypermedia
De-sign with OOHDM. In Proceedings
of the ACM International Confer-ence on Hypertext, Hypertext ’96, Washington.
[UnS87] D. Ungar, R. B. Smith (1987). Self:
The power of Simplicity, In:
OOPSLA’87 Proceedings, pp. 227-242.
[W3C99] World-Wide Web Consortium (1999). XML: eXtensible Markup
Language. http://www.w3.org/XML
[Zhao97] Y. Zhao (1997). A Single Login Web
Service Integrator - WebEntrance.
In: Computer Networks and ISDN Systems 29, Special Issue on the 6th Intl. World-Wide Web Conference, Santa Clara, CA, USA.
About the Authors
Guntram Graef obtained a Master degree in computer science from the University of Karlsruhe in
1998. In 1999 he joined TecO as a research assistant and works in collaborative Web engineering projects. Former to that he was an independent consultant designing distributed applications for major companies in Europe and Asia. His current research interest is in methodologies and tools for large-scale Web development, e-commerce system architectures, user profiling and applications of XML.
Martin Gaedke is research assistant at the Telecooperation Office (TecO) of the University of
Karlsruhe and technical lead in collaborative Web engineering projects. His research interests is in application of software engineering practice to applications in the WWW, and specifically in com-ponent-based software engineering, reuse, frameworks, reflection, software architectures, patterns and software evolution for Web applications. He obtained a Master degree in computer science from the University of Karlsruhe in 1997.
