Metamodeling Architecturesand Interoperability of

Web-enabled Information Systems

Marie-Noelle Terrasse, George Becker, and Marinette SavonnetLaboratoire LE2I, Universite de Bourgogne

B.P. 47870, 21078 Dijon Cedex, FranceE-mail: {terrasse,becker,savonnet}@khali.u-bourgogne.fr

January 26, 2004

1 Introduction

Web-enabled information systems are particular challenges in interoperability in thesense that they are strongly heterogeneous in terms of their application domain, as wellas of their modeling languages. Our approach is to use metamodeling environments forweb-enabled information systems. Metamodeling is extensively used in informationsystem analysis & design as a powerful mechanism for abstraction. This abstractionenables modelers to describe complex systems, e.g., systems that provide many dif-ferent user-specific functionalities within complex services (including distribution ofresources, concurrency, etc.). Furthermore, the design phase itself is being carried outunder demanding requirements, e.g., making software reuse possible, providing formaltoolboxes for simulation and validation of models, and automatic generation of codefrom formal specifications. In such a demanding context, abstraction is carried out bytwo complementary mechanisms: abstraction by conceptualization and abstraction byprojection. An overview of such abstraction mechanisms follows.

Abstraction by conceptualization Abstraction by conceptualization strives to struc-ture a given description of an information system in several layers that constitute ametamodeling architecture. Many modeling environments refer to the OMG’s meta-modeling architecture [UML, 2000], depicted in Figure 4, which goes beyond the two-layer description of databases by proposing four layers: instance, model, metamodel,and meta-metamodel layers.

• The metamodel layer describes –at an abstract level– an application domain.More precisely, the metamodel layer defines which modeling constructs are tobe used for modeling of all information systems belonging to a given applicationdomain. For example, the profiles of the OMG are standard extensions of theUML metamodel that are dedicated to specific application domains.

1

[Selic and Rumbaugh, 1998] propose three main constructs capsule, port, andconnector for their extension of UML to real-time modeling. They explain theneed for a new construct port: “Although ports are boundary objects that actas interfaces, they do not map directly to UML interfaces. A UML interface ispurely a behavioral thing –it has no implementation structure. A port, on theother hand, includes both structure and behavior.”

• The meta-metamodel layer describes how the real world is seen. A meta-meta-model encompasses a high-level description of the underlying logics (e.g., boo-lean or modal logics), a particular time model or spatial model which will beused to capture the semantics of the real world. For example, the TAU Tem-poral Object Model [Kakoudakis, 1996] (which defines a time axis as discrete,linear, totally ordered and bounded at both ends) is used for the TUML defini-tion [Svinterikou and Theodoulidis, 1997]. [Price et al., 1999] propose “a rangeof different semantics and models for space time, and change processes”.

Abstraction by projection Abstraction by projection relies upon the principles ofseparation of concerns [Breu et al., 1998] [Mili et al., 1995] [Schatz and Huber, 1999]and combination of concerns [Bezivin, 1998]. Separation of concerns is implementedthrough a multi-view model which encompasses orthogonal views of the system. Aspointed out by [Geisler et al., 1998], different views of such a model are syntacticallyindependent yet they are semantically coupled. For example, in a UML-based meta-modeling architecture, a metamodel provides modelers with a multi-view model of theinformation system which encompasses 9 diagrams for description of the system:

• The user view (Use Case Diagram) describes the system as a set of main func-tionalities (called use cases) that are offered to different types of users (calledactors). Generally, the user view outlines –through traceability information– re-lationships between functionalities and classes.

• The structural view (using Class and Collaboration Diagrams) is the core ofthe static description of the system. The Class Diagram presents the systemin terms of entities and their “semantically strong” relationships and depen-dencies (i.e., conceptual dependencies that are noticeable between instances,such as compositions or aggregations, as well as contextual dependencies thatare due to signatures of methods). The Collaboration Diagram describes “se-mantically weak” relationships and dependencies such as roles between entities[Lyons, 1998] [Selic and Rumbaugh, 1998], as well as operational dependencies(i.e., dependencies that are due to bodies of methods: a method uses a localobject belonging to another class [Desfray, 1997]).

• The behavioral view (Sequence, Activity, and StateCharts Diagrams) describesthe dynamics of the system. These diagrams establish (for a single class or forseveral classes) protocols through which objects communicate and interact. TheSequence Diagram is dedicated to the description of purely sequential object-level activities, while the Activity Diagram describes the main patterns of com-plex internal interactions (including synchronization of parallel actions). The

2

StateCharts Diagram formally expresses behavior of a class in terms of a statemachine, which may facilitate formal verification.

• The physical view (Components and Deployment Diagrams) describes the sys-tem in terms of a set of physical components (data, code, and documentationmodules) which are located on hardware nodes and related by physical links,i.e., network connections.

• The instance view (Object Diagram) presents the system in terms of related ob-jects. The Object Diagram is a snapshot of the system as well as an instantiationof the Class Diagram.

Those orthogonal views have to be combined in order to provide a consistentspecification of the system, which is called combination of concerns. Combinationof concerns must guarantee that views are consistent which each other and that eachpiece of information appears in at least one of the proposed views. In order to verifythe meaningfulness of multi-view models, two main approaches have been proposed.The first one [Falcon et al., 1999] [Pons et al., 1998] is an “a posteriori” approach thatchecks each model separately. The metamodel implements separation of concerns,(i.e., defines a set of views to be used) but not combination of concerns. In the sec-ond approach, which is used in many metamodeling architectures [Breu et al., 1997][Clark and Evans, 1998] [Cook et al., 1999] [France and Rumpe, 1999], the metamodelof the modeling language implements both principle of separation and that of combi-nation of concerns: the metamodel defines a set of views to be used and guarantees thatthe combination of those views will be meaningful.

Overview of the chapter Through the above abstraction mechanisms, metamodelsappear to be a major tool for analysis & design of information systems. Metamodelingarchitectures are becoming the backbone of modeling environments in which model-ing is carried out through a two-stage process: first, defining a convenient metamodelfor the application domain, and second, describing the model of the application as aninstantiation of the application-domain metamodel. We propose to use metamodel-ing –and its abstraction mechanisms– for modeling of and achieving interoperabilityof web-enabled information systems. Abstraction by projection allows us to describeuser-specific features of web-enabled information systems and their security aspectsmore accurately. Abstraction by conceptualization allows us to define abstract basesof agreement for interoperability of web-based information systems. In Section 2 wepresent a two-fold point of view on web-enabled information system development (e.g.,analysis & design and interoperability of information systems) from which we derivea structure for web-enabled frameworks. Section 3 presents our UML-based proposalfor various components of such a framework. Section 4 discusses future developmentsof metamodeled-web-based information systems.

3

Coreelaboration

&documentation

Consistent and completespecification

Specification Toolboxvalidation & refinement

Metamodel&

Model

Coreelaboration

&documentation

Metamodel&

Model

Programming Toolbox

coding & testing

Code

a) Formal environment b) Programming environment

Figure 1: Formal versus programming modeling environments

2 Context for web-enabled information systems

Web-enabled information systems lie at the crossroad of two major trends of softwareengineering: analysis & design and interoperability. They are presented in Section 2.1and Section 2.2, respectively. From these two trends, we derive our perspective onweb-enabled information systems in Section 2.3.

2.1 Modeling environments for analysis & design

The increasing interest in UML made it possible to develop a large number of modelingenvironments, both in academia and in industry. These environments rely upon a corecomponent which is dedicated to elaboration of metamodels and models (descriptionof different views, editing of documentation, etc.). Based upon the overall objectives ofan environment, additional toolboxes are proposed. We distinguish two main familiesof modeling environments.

• Formal-oriented environments, see Figure 1.a, propose specification toolboxeswhich provide modelers with formal tools for validation and refinement of mod-els. Validation of a model is essential in the sense that a formal specification (i.e.,a specification complying with a formal syntax) may be incomplete or inconsis-tent. There is no real difference in quality between an informal specification anda formal but wrong specification. Formal tools allow modelers to guarantee thecompleteness and consistency of a formal specification [Clarke and Wing, 1996][Ober, 2000]. Furthermore, formal toolboxes make modeling environments moremature by making new tools, e.g., for verification [Evans, 1998] and test gener-ation [Ammann and Black, 1999], available to modelers.

For example, the SRC environment [Heitmeyer et al., 1998] uses model check-ing for analysis of software requirements and production of specification. Their

4

Coreelaboration

&documentation

Specification

Specification Toolboxvalidation & refinement

Metamodel&

Model

Programming Toolbox

coding & testing

Code

(1) (2)

Figure 2: Comprehensive modeling environments

formal work on underlying algorithms has allowed them to guarantee quality ofsuch specifications [Jeffords and Heitmeyer, 2001].

• Programming-oriented environments, see Figure 1.b, propose programming tool-boxes which provide modelers with tools for generation of code (e.g., C++ orJava) from a model of the system. Most of the commercial products for UML-based modeling include such toolboxes. In many cases, code generation is car-ried out from an unverified model: an error in the model is propagated into thecode.

For example, [Nordstrom et al., 1999] have proposed a UML-based environment,called Model Integrated Computing (MIC), for generation of programs. MICencompasses a partial validation of the model. They have further extended MICinto a Generic Modeling Environment [Ledeczi et al., 2001].

The evolution of major modeling environments seems to inherently proceed to-wards the integration of formal and programming tools, see Figure 2. Modelers usethe core component of modeling environments for defining a metamodel and a model.Then, they use formal tools in order to produce a correct specification from their modelof an application. Finally, programming tools translate the specification into efficientcode.

2.2 Interoperability of information systems

With recent developments in interoperability of information systems, a priori solutionsare preferred: interoperations are no longer constructed “on the spot” by integration

5

of participating information systems. The common knowledge is pre-determined andaccepted by all participants: it is a basis of agreement. Each information system ofthe interoperation group communicates with other systems by matching its own infor-mation against the basis of agreement. Bases of agreement can have very differentstructures: a common vocabulary or taxonomy, a common model, common protocols,etc. Two major trends arise in building bases of agreement:

• Domain-based solutions have become common. They identify the common knowl-edge that is shared by all applications in a given domain, e.g., [Guarino, 1995]proposed to express such a common knowledge in terms of part-whole relationand connection relation, called mereology and topology, respectively. Applica-tion domain descriptions are either picked up from a collection of existing de-scriptions or are produced from scratch. Such solutions comply with the reuserequirement that developed with Object Orientation. Analysis & design itselfevolves towards more general environments whose objectives are no longer re-stricted to producing a “private” model for a single application.

Similarly, interoperability is built upon a basis of agreement generally related toan application domain. The SHOE project on Web ontology [Heflin, 2001] is anexample of such a domain-based approach. For a survey of the state-of-the-artof domain-based approaches, see [Leclercq, 2000] or [Jouanot et al., 2000].

• Metamodel-based solutions identify main concepts of each modeling language,e.g., O.O. classes and associations, relational tables and keys, etc. They or-ganize these concepts into a common structure, e.g., an inheritance hierarchy[Lou, 1997], graph [Atzeni and Torlone, 1994], lattice [Jeusfeld and Johnen, 1994][Nicolle, 1996]. Such approaches are very appropriate as long as the set of con-cepts is stable enough: the common structure is defined first and then formaltools –depending on the chosen structure– are applied in order to perform modeltranslations. One of the major problems of such approaches is making their struc-ture evolve in order to integrate new concepts. Alternative proposals organize themetamodels themselves by using the category theory [Frederiks et al., 1997] orhierarchy structures [Falkenberg and Han Oei, 1994].

UML-metamodeling is the first step towards reintegration of the above two trendssince it provides an integrated architecture for expressing domain classes and meta-model constructs. For example, the OMG’s profiles [OMG, 1999] encompass bothmetamodel-level constructs and model-level classes. In the following section, wedefine an integrated point of view on analysis & design and interoperability of web-enabled information systems that is based on a UML-metamodeling architecture.

2.3 Consequences for web-enabled information systems

Web-enabled information systems require efficient and integrated solutions for analysis& design and interoperability problems. Thus, we propose an integrated environmentin which the “formal versus programming” structure of analysis & design is extendedtowards interoperability. This integrated environment, depicted in Figure 3, encom-passes 5 components:

6

Metamodel&

Model

Coreelaboration

&documentation

InteroperabilitySpecification

FormalInteroperationToolbox

ApplicationSpecification

SpecificationToolbox

ProgrammingToolbox

Code

AgreementToolbox

Bases ofAgreement

Figure 3: Modeling environment for interoperability

• The core component is dedicated to elaboration of models and their documen-tation. This core component complies with the metamodeling approach: firstdefining a metamodel (for description of the application domain of informationsystems or the context of interoperability), then describing the model.

• The specification toolbox encompasses tools for formal operations on models:model checking, validation of models against users requirements, refinement ofmodels, etc. By using those tools, modelers can produce a reliable specificationof an information system.

• The programming toolbox is used for generation and testing of executable codebased upon the specification of an information system.

• The formal interoperation toolbox produces an abstract basis of agreementfrom a set of specifications of the interoperating information systems. In Sec-tion 3.3 we present the construction of such abstract basis of agreement in moredetails.

• The agreement toolbox is used for generation of actual bases of agreement fromabstract bases of agreement.

The following section presents the main features of an UML-based integrated en-vironment for management of web-enabled information systems.

7

3 UML-metamodeling and web-enabled information sys-tems

Several authors [de Lima and Price, 1998] [Baumeister et al., 1999] [Koch et al., 2000]have proposed UML extensions for the Web. Most of them provide modelers with newconcepts for describing navigation. Yet, it is possible to go further by using UMLmetamodeling architectures. Section 3.1 emphasizes advantages of abstraction by pro-jection for defining core components of modeling environments. Sections 3.2 and 3.3present our approach to specification toolboxes and formal interoperation toolboxes,respectively.

3.1 Our approach to core components

Our UML-based core component allows designers to improve web-enabled models byexploiting different views of UML models.

• The user view is a major tool for providing various users with their specific viewsof the system. The Use Case Diagram establishes which type of users is autho-rized to use each functionality. According to the Unified Development Process[Jacobson et al., 1999] use cases are first described and then used as a basis fordetermining the Class and Sequence diagrams of the UML model. By using thetraceability information (between use cases and classes), it is easy to establishwhich classes are necessary for each type of users. Such information may beused for validating (or defining) user-specific views of the information system.

• The behavioral view of information systems (Sequence, Activity, and State-Charts Diagrams) is becoming increasingly important for web-enabled informa-tion systems. Since users of these systems are no longer limited to local andknowledgeable users, it is necessary to provide the users with a precise descrip-tion of operations that can be activated. Many authors point out the problem ofreadability of behavioral descriptions for a casual user. Some authors proposealternatives, e.g., the “light-weight tools” of Heitmeyer & al. [Heitmeyer, 1998]whose users do not need “advanced mathematical training and theorem provingskill”.

• The physical view of a system (Component and Deployment Diagrams) has twomajor applications. First, the description of physical components of a system,combined with the localization of these components on different sites (host ma-chines), allows modelers to describe distribution of information. Second, theDeployment Diagram which describes types of links between the sites can becombined with models of interactions between system components in order toprovide security specialists with information needed for design of security tools.See, for example, Jurjens’ extension of the UML [Jurjens, 2001] for developmentof secure systems.

8

Meta-metamodel Layer: L3

Metamodel Layer: L2

Model Layer: L1

Instance Layer: L0

UML as a

descriptive

language

Meta-

metamodel

. . . Metamodel . . . Metamodel . . . Metamodel . . .

. . . Model . . . Model . . . Model . . .

. . . Instance . . . Instance . . . Instance . . .

Figure 4: UML-based four-layer metamodeling architecture

3.2 Our approach to a specification toolbox

Our specification toolbox relies upon a UML-based metamodeling architecture whichcomplies with the OMG’s architecture, as depicted in Figure 4. In order to make ournotation more explicit, in the following discussion we denote by L0, L1, L2, and L3 theinstance, model, metamodel, and meta-metamodel layers, respectively. Consistentlywith this notation, we will attach a superscript (from 0 to 3) to each element that islocalized on the corresponding layer of the metamodeling architecture.

We reorganize the two uppermost layers of the metamodeling architecture into amirroring structure that consists of abstract descriptions (called modeling paradigms),possibly mixing several different languages, as well as concrete uni-language descrip-tions that are specializations of the UML metamodel. Both layers, as well as theirrelationship, are described below.

Meta-metamodel layer Our meta-metamodel layer comprises modeling paradigmsthat describe –in terms of concepts that are interrelated by constraints– the semantics ofthe real world. As defined in [Terrasse, 2001], a modeling paradigm mp is described–using the English language, logics and the set theory– by two sets, El3(mp) andC3(mp). The set El3(mp) contains descriptions of elementary concepts, while the setC3(mp) contains constraints among the concepts of El3(mp).

Our first example is the general O.O modeling paradigm, named gmp, which ap-pears in the UML approach [UML, 2000]. El3(gmp) contains concepts object (withidentity, state, and behavior), class, generalization, etc. C3(gmp) contains constraintslike each object belongs to a class.

Another example relates to an architecture description language, named C2-ADL.[Medvidovic and Rosenblum, 1999] describe the C2-style architecture in English: “co-nnectors transmit messages between components, while components maintain state,perform operations, and exchange messages with other components via two interfaces

9

(named ”top” and ”bottom”). ... Inter-component messages are either requests for acomponent to perform an operation or notifications that a given component has per-formed an operation or changed state”. Thus, as depicted in Figure 5, El3(mp C2)contains concepts connector, component, interface, message, etc. The authors alsogive constraints which belong to C3(mp C2), such as “components may not directlyexchange messages; they may only do so via connectors”. A formal description of C2,in language Z, is given in [Medvidovic et al., 1996].

Modeling paradigms may use a various number of concepts; each of them beingdescribed with more or less detail. For example, a modeling paradigm may use aunique concept of class, while another modeling paradigm may use different conceptssuch as interface, abstract class, and implementation class. Thus, we define a partialordering relationship between modeling paradigms as follows: a modeling paradigmmp1 is subsumed by a modeling paradigm mp2, which is denoted by mp1 � mp2, ifwe have both extended inclusion of concepts1 and subsumption of constraints2.

Two modeling paradigms that are related by � or by the inverse relationship sub-sumes (denoted by �) are said to be comparable. We are interested in all model-ing paradigms that are subsumed by the general modeling paragraph gmp, as pre-sented below. Let us call by RestrictMP the set of such modeling paradigms. Ourmeta-metamodel layer may be structured as a poset of modeling paradigms, whoseRestrictMP determines a sub-poset in which any modeling paradigm is subsumed bygmp.

Metamodel layer Our objective is to build our metamodel layer as a mirror of theposet of modeling paradigms: the generic modeling paradigm gmp is instantiated intothe UML metamodel itself, and other modeling paradigms are instantiated into spe-cializations of the UML metamodel. Analogously to modelers’practice that extendsUML metamodel to various application domains (see [Herrero et al., 2000] for a de-tailed example) we use tailoring mechanisms of UML in order to instantiate modelingparadigms as metamodels. These tailoring mechanisms are constraints (written in theOCL language), tag values (extra components with pre-defined values which are at-tached to an existing UML-construct) and stereotypes (which use constraints and tagvalues for modification of a construct in order to create a new construct).

An instantiation function E3,2 is defined –from RestrictMP to a set of UMLmetamodels– in order to build metamodels from modeling paradigms. Let us considera modeling paradigm mp. E3,2 associates each concept of El3(mp) with one or moreelementary components of the UML language. These components are either standardUML constructs or stereotypes which include some of the constraints of C3(mp). Thus,we assume that mp’s corresponding metamodel mm = E3,2(mp) is described by a setof elementary components El2(mm) and a set of constraints C2(mm). C2(mm) con-tains instantiations of some constraints of C3(mp), and additional constraints due tothe instantiation process itself.

[Robbins et al., 1998] propose an extension of UML for C2-ADL which corre-sponds with the example given above. They define C2-interface as stereotype of theUML interface with a tagged value (top, bottom). C2-request and C2-notificationare defined as stereotypes of the UML operation with a constraint to forbid any re-

10

turn value and having tag values to distinguish requests from notifications. Similarly,C2-connector and C2-component are defined as stereotypes of the UML class, and astereotype of the UML association is used to attach C2-component and C2-connector.El2(mm C2) is depicted in Figure 5 and C2(mm C2) contains instantiations of con-straints of C3(mp C2).

Meta-metamodel layer

Metamodel layer

gmp

mp2concept:

weak encapsulation

mp6

concept:

synchronization

mp3

concept:

time-models

mp5

constraint:unique time-model

mp7

concept:

spatial-model

mp4concept:

C2-ADL

El3(mp4)={connector,component,

interface,

message, . . .}

C3(mp4)={connector constraint,. . .}

��

�

�

� �

�

UML Metamodel

mm2

mm6

mm3

mm5

mm7

mm4 El2(mm4)={C2-connector,C2-component,

C2-interface,

C2-notification,

C2-request, . . .}

C2(mm4)={. . .}

Figure 5: Our mirroring structure

A two-fold mirroring structure We require that our instantiation complies with theordering of modeling paradigms so that our metamodel layer can mirror the meta-metamodel layer. An instantiation is said to be fully compliant if each modelingparadigm is mirrored by its instantiation as a metamodel and each ordering relation-ship between modeling paradigms is mirrored by an inheritance relationship betweenmetamodels. Figure 5 presents an example of a fully compliant mirroring structure inwhich multiple inheritance is necessary (observe the construction of mm6 from mm2

and mm5).As a consequence, we have to instantiate all modeling paradigms, even the am-

biguous ones. In the following, we distinguish two types of ambiguities: initial-ambiguities which originate in the UML metamodel itself (see [Evans et al., 1998a]),

11

and extension-ambiguities due to ambiguous extensions.In Figure 5, the modeling paradigm mp3 is ambiguous since it includes several pos-

sible time-models [Price et al., 1999]. The modeling paradigm mp5, which is limitedto one of Price’s time-models can be unambiguous (if it contains no initial-ambiguity).

We say that mp is a consistent modeling paradigm if the instantiated set of con-straints C2(mm) is consistent. We say that mp is an unambiguous modeling paradigmif the instantiated set of constraints C2(mm) contains no initial- or extension- ambigu-ity. See [Terrasse and Savonnet, 2000] for more details on the mirroring structure andformalization issues.

3.3 Our approach to a formal interoperation toolbox

Formal interoperation toolboxes aim at producing an abstract basis of agreement fromwhich an actual basis of agreement can be obtained by instantiation. Our strategy con-sists of constructing an abstract basis of agreement in terms of a metamodel. As longas possible, the inheritance hierarchy of metamodels is used for such a construction.Since ancestors of a given metamodel are more general (they do not support that manyextensions, and they are not subjected to that many constraints), it is likely that a com-mon ancestor of two given metamodels is a good basis of agreement. For example, inFigure 5, the common ancestor of mm6 and mm7 (i.e., mm5) supports their commontime extensions, yet it is independent of their specific extensions for spatial informationand synchronization. As pointed out above, such a common ancestor can be ambigu-ous: it is thus possible that inconsistent choices were made in its descendants. We haveto guarantee that the metamodel proposed as a basis of agreement is not ambiguous.In case of failure, i.e., if no unambiguous metamodel can be extracted from the inher-itance hierarchy, we use the meta-metamodel level to build a modeling paradigm foragreement.

Let us consider two information systems IS1 and IS2 and their respective meta-models, mm1 and mm2. We use a two-fold strategy to define an underlying abstractbasis of agreement. See [Terrasse et al., 2001] for a precise presentation of formal op-erations on metamodels and modeling paradigms.

• At the metamodel level, by using the inheritance hierarchy of metamodels, wesearch for the first common ancestor mm of mm1 and mm2. If that first com-mon ancestor mm is an unambiguous metamodel, it is used as an abstract basisof agreement. If mm is ambiguous, we search for two intermediate and consis-tent metamodels that can be integrated. Their integrated metamodel is used asan abstract basis of agreement.

As depicted in Figure 6, in which large dark circles indicate bases of agreement,there are the following possibilities:

– Agreement on a relaxed metamodelSee part (a) of Figure 6. Let us denote by mm the first common ancestorof mm1 and mm2 in the inheritance hierarchy. The metamodel mm corre-sponds to weaker constraints than those of both mm1 and mm2 metamod-

12

Meta-metamodel Layer: L3

Metamodel Layer: L2

(a): Relaxed metamodel (b): Integrated metamodel (c): Instantiated metamodel

mm

mm1 mm2

mm

mm1 mm2

mm′

1mm′

2

mm

mm1 mm2

mm′

1mm′

2

�

E3,2

mp

E3,2

mp′

Figure 6: Strategies for building abstract bases of agreement

els. If mm is an unambiguous metamodel, mm supports the interoperabil-ity of mm1 and mm2.

– Agreement on an integrated metamodelSee part (b) of Figure 6. If the common ancestor mm is an ambiguousmetamodel, let us consider mm′

1and mm′

2, the first unambiguous meta-

models on the path from mm to mm1 and from mm to mm2, respectively.If mm′

1and mm′

2are consistent, then it is possible to integrate them in or-

der to construct a metamodel for agreement. The integrated metamodel isbuilt by restricted union of concepts3 and restricted union of constraints4.

• If it is impossible to build an integrated metamodel, then we search –at themeta-metamodel level– for a suitable modeling paradigm to be instantiated asan abstract basis of agreement. Thus, we have agreement on an instantiatedmetamodel (see part (c) of Figure 6). If mm′

1and mm′

2are not consistent with

each other, it is impossible to integrate mm′

1 and mm′

2 into a metamodel. Thesolution is to move up to the meta-metamodel layer. Let us denote by mp themeta-metamodel corresponding to the ambiguous metamodel mm. We chooseone of the unambiguous modeling paradigms from descendants of mp, and wedenote it mp′. The instantiated metamodel E3,2(mp′) is used as an abstract basisof agreement.

Since the resulting metamodel may be semantically distant from one (or both)of the given metamodels, we have to define a strategy for choosing among un-ambiguous heirs of mp. Towards this end, we intend to utilize distance measurebetween metamodels [Terrasse, 2001] which is computed as a weighted sum of

13

elementary distances between corresponding elements of metamodels (i.e., cor-responding components and corresponding constraints). Elementary distances,as well as weights used for their combination, can be fine-tuned according to thesemantics of application domains.

From this construction process, it appears that construction of abstract bases ofagreement relies upon the structure of modeling paradigms within their poset. Qualityof relaxed and integrated metamodels depends on the inheritance hierarchy of meta-models which is –by our full compliance requirement– a mirror of the poset structure.Quality of instantiated metamodels depends directly on the poset structure of modelingparadigms.

4 Conclusion

As pointed out by [Koch et al., 2000b], there is a consensus forming against “the clas-sical approach of integrating all sources against a single common ’global’ informationmodel”. Thus, metamodeling environments appear to be a winning proposition. Nev-ertheless, a major issue for metamodeling environments is to evaluate to what degreethe modelers are able to modify their “way of modeling” [Wijers, 1991] in order tointegrate metamodeling into their modeling process. Below, we discuss two aspects ofthis issue.

A new concept of common knowledge As pointed out by [SCSC, 1999], inter-operability of information systems has to cope with the “bases of agreement versuspre-existing knowledge” problem: “A key issue is the nature of information. With-out a shared context and common foundations of understanding there is no infor-mation -just a representation with no meaning shared between people”. For a longtime, database federation research has been exhibiting “boundary examples” wherethe commonly-accepted pre-existing assumptions are no longer valid. Since the “ba-sic system of reference is shifting from real organizations to virtual communities ofinterest” [Bodart, 1999], the notion of a semantic domain with delimited boundariesno longer exists. Thus, the assumption of pre-existing knowledge appears as a purelyhypothetical point of view. This emphasizes the role of bases of agreement and, as aconsequence, we are caught in a dead-end situation.

According to [SCSC, 1999], the solution might lie elsewhere: “There is, post-modernists would argue, no absolute foundation of our reality. Perhaps there is in-stead, elemental constructs in our way of thinking, the constructs that enable us todiscern, interpret and understand information”. We believe that the use of abstractbases of agreement opens such a new perspective. Let us present an example: patternsfor graph-dynamics have been proposed [Benslimane et al., 2000] [Zhou et al., 2000].They describe virtual networks in terms of a collection of functionalities. Each actualnetwork is derived from a pattern by choosing a set of relevant functionalities. Such“patterns of dynamics” can be encapsulated –at the meta-metamodel level– as a globalconstruct representing the common understanding of graph-based navigation.

14

A new role for domain experts Using our proposal for metamodeling environments,domain experts will have to take charge of defining specific metamodels for each ap-plication domain, tuning algorithms for computing distances between metamodels,validating the poset of modeling paradigms, etc. All these tasks seem to be remotefrom the classical roles of domain experts, which we discuss below. One one hand,[Opdahl, 1996] pointed out the need for information system architectures that can “fitwith the enterprise’s organization structure, culture, organizational centralization de-grees and philosophy”, and he developed a proposal for IS-architecture alignment (withrespect to enterprises) in the RAISA project [Opdahl, 1998]. In such a context, expertsno longer need to construct ontologies by putting together “a high density of knowledge... with a comparably small number of concepts” [Koch et al., 2000a], but they needto work at the knowledge management level. On the second hand, under the pressureof model evolution research, domain experts have to establish relationships betweensemantically close descriptions of the real world. Thus, a new role for domain expertsis emerging. First, they are supposed to express a domain-specific point of view onthe real world in terms of abstract concepts that will be later translated into modelingconcepts. Second, they are supposed to validate evolution of these modeling conceptsin order to tune the domain description (for a different but close domain or for the samedomain at another point of time).

At the first glance, introduction of metamodeling for web-enabled information sys-tems appears to be a paradox. On one hand, metamodeling is very demanding in termsof a modeling framework, a modeling process, modelers’ practice, experts’ involve-ment, etc. That many requirements can make metamodeling unsuitable in the contextof web-enabled information systems. On the other hand, metamodeling is aligned withthe actual evolution of information technology,i.e., towards more abstract and moreformal methods, towards integrated frameworks, towards involvement of high-levelspecialists. Since it is likely that web-enabled information systems will have to relyupon high-level technology in order to survive, the paradox is probably only apparent.

AcknowlegmentsOur approach to metamodeling originate in work that has been carried out at the Na-tional Institute of Standard and Technology with Christopher E. Dabrowski, LeonardJ. Gallager, and Lisa J. Carnahan. Many thanks to them for their help during the initialefforts in developing our perspective on metamodeling.

15

Notes

1Extended inclusion of concepts means that each concept of El3(mp2) is either amember of El3(mp1) or a generalization of a concept of El3(mp1), where a generalizedconcept may have fewer features than its specialized concept has.

2Subsumption of constraints means that by using C3(mp1) as a hypothesis, it ispossible to prove that each constraint of C3(mp2) holds.

3Restricted union of concepts means that each concept of two given sets eitherappears in the union or its generalization belongs to the union.

4Restricted union of constraints means that by using the union as an hypothesis, itis possible to prove that each constraint of two given sets holds.

References

[Ammann and Black, 1999] Ammann, P. and Black, P. E. (1999). Abstracting FormalSpecifications to Generate Software Tests via Model Checking. In Proceedingsof the 18th Digital Avionics Systems Conference, DASC’99, Saint Louis, Missouri,USA. Available at URL www.nist.gov.

[Atzeni and Torlone, 1994] Atzeni, P. and Torlone, R. (1994). Managing Heteroge-neous Data Models in a Graph-Theoretic Framework. In Proceedings of the 3rdInternational Conference on Information and Knowledge Management, CIKM’94,Gaithersburg, Maryland, USA.

[Baumeister et al., 1999] Baumeister, H., Koch, N., and Mandel, L. (1999). Towardsa UML Extension for Hypermedia Design. In UML’99 - The Unified ModelingLanguage (Beyond the Standard), pages 614–629. Springer Verlag, LNCS 1723.Available at URL http://helios.pst.informatik.uni-muenchen.de/bibtex.

[Benslimane et al., 2000] Benslimane, D., Leclercq, E., Savonnet, M., Terrasse, M.,and Yetongnon, K. (2000). On the Definition of Generic Multi-Layered Ontologiesfor Urban Applications. In Computers, Environment and Urban Systems, volume 24,pages 191–214. Elsevier. Special Issue on Knowledge Engineering in Urban Plan-ning.

[Bezivin, 1998] Bezivin, J. (1998). On Different Interoperability Modes in SoftwareEngineering: the Case of Modeling Activities at OMG. In Proceedings of SoftwareEngineering’98, Paris, France.

[Bodart, 1999] Bodart, F. (1999). Challenging aspects about the evolution of infor-mation systems. In Proceedings of the Information Systems Foundations WorkshopOntology, Semiotics and Practice, Lausanne, Switzerland.

16

[Breu et al., 1998] Breu, R., Grosu, R., Huber, F., Rumpe, B., and SCHWERIN, W.(1998). Systems, Views and Models of UML. In Schader, M. and Korthaus, A.,editors, The Unified Modeling Language, Technical Aspects and Applications, pages93–109. Physica Verlag. Available at URL http://www.cs.york.ac.uk/puml.

[Breu et al., 1997] Breu, R., Hinkel, U., Hofmann, C., Klein, C., PAECH, B., Rumpe,B., and Thurner, V. (1997). Towards a Formalization of the Unified Modeling Lan-guage. In Proceedings of ECOOP’97, pages 344–366. Springer Verlag, LNCS 1241.Available at URL http://www4.informatik.tu-muenchen.de/papers.

[Clark and Evans, 1998] Clark, A. and Evans, A. (1998). Foundations of the Uni-fied Modeling Language. In 2nd Northern Formal Methods Workshop, Ilk-ley, Electronic Workshops in Computing. Springer Verlag. Available at URLhttp://www.cs.york.ac.uk/puml.

[Clarke and Wing, 1996] Clarke, E. M. and Wing, J. M. (1996). Formal Methods:State of Art and Future Directions. In ACM Computing Surveys, 28(4), pages 626–643. ACM.

[Cook et al., 1999] Cook, S., Kleppe, A., Mitchell, R., Rumpe, B., WARMER, J., andWills, A. C. (1999). Defining UML Family Members Using Prefaces. In Mingins,C. and Meyer, B., editors, Proceedings of Technology of Object-Oriented Languagesand Systems, TOOLS 32, pages 102–114. IEEE.

[de Lima and Price, 1998] de Lima, F. A. and Price, R. (1998). Defining UMLStereotypes for Modeling Web Information Systems. Available at URLhttp://www.inf.ufrgs.br/vflima/pro.html.

[Desfray, 1997] Desfray, P. (1997). Modelisation par objets - La fin de la programma-tion. InterEditions. See also Object Engineering: the Fourth Dimension, AddisonWesley, 1994.

[Evans, 1998] Evans, A. (1998). Reasoning with UML Class Diagrams. In Proceed-ings of the 2nd IEEE Workshop on Industrial Strength Formal Specification Tech-niques, pages 102–113. IEEE. Available at URL http://www.cs.york.ac.uk/puml.

[Evans et al., 1998a] Evans, A., Bruel, J.-M., France, R., Lano, K., and Rumpe,B. (1998). Making UML Precise. In OOPSLA’98 Workshop on “FormalizingUML. Why and How?”, Vancouver, Canada. OOPSLA’98. Available at URLhttp://www4.informatik.tu-muenchen.de/papers.

[Falcon et al., 1999] Falcon, P., Laleau, R., Mammar, A., and Polack, F. (1999). For-mal Specification of the UML Metamodel for Building Rigorous CAiSE Tools.Technical Report CEDRIC.

[Falkenberg and Han Oei, 1994] Falkenberg, E. D. and Han Oei, J. (1994). MetaModel Hierarchies from an Object-Role Modeling Perspective. In Proceedings ofthe 1st International Conference on Object-Role Modeling, ORM-1, Magnetic Is-land, Australia.

17

[France and Rumpe, 1999] France, R. and Rumpe, B. (1999). UML’99 - The UnifiedModeling Language (Beyond the Standard). Springer Verlag, LNCS 1723. Availableat URL http://www4.informatik.tu-muenchen.de/papers.

[Frederiks et al., 1997] Frederiks, P., ter Hofstede, A., and Lippe, E. (1997). A Unify-ing Framework for Conceptual Data Modelling Concepts. Information and SoftwareTechnology, 39(1), pages 15–25.

[Geisler et al., 1998] Geisler, R., Klar, M., and Pons, C. (1998). Dimensions and Di-chotomy in Metamodeling. Technical Report TR 98-5, TU Berlin, FachbereichInformatik. Available at URL http://tfs.cs.TU-berlin.de.

[Guarino, 1995] Guarino, N. (1995). Formal Ontology, Conceptual Analysis andKnowledge Representation. International Journal of Human and Computer Studies,43.

[Heflin, 2001] Heflin, J. (2001). Towards the Semantic Web: Knowledge Represen-tation in a Dynamic, Distributed Environment. Phd thesis, University of Maryland,College Park.

[Heitmeyer, 1998] Heitmeyer, C. (1998). On the Need for Practical Formal Methods.In Proceedings of the 5th International Symposium on Formal Techniques in Real-Time and Real-Time Fault-Tolerant Systems, FTRTFT’98, Lyngby, Denmark, pages18–26. Invited paper, available at URL chacs.nrl.navy.mil/SCR.

[Heitmeyer et al., 1998] Heitmeyer, C., Kirby, J., Labaw, B., and Bharadwaj, R.(1998). SCR: A Toolset for Specifying and Analyzing Software Require-ments. In Proceedings of the 10th Annual Conference on Computer-AidedVerification, CAV’98, pages 526–531, Vancouver, Canada. Available at URLchacs.nrl.navy.mil/SCR.

[Herrero et al., 2000] Herrero, J. L., Sanchez, F., Lucio, F., and Bonilla, M. T. (2000).Changing UML Metamodel in Order to Represent Concern Separation. ECOOP’00Workshop 14 on Defining a Precise Semantics for UML, Sophia Antipolis, France.

[Jacobson et al., 1999] Jacobson, I., Booch, G., and Rumbaugh, J. (1999). The UnifiedSoftware Development Process. Addison-Wesley. ISBN 0-201-57169-2.

[Jeffords and Heitmeyer, 2001] Jeffords, R. D. and Heitmeyer, C. (2001). An Algo-rithm for Strengthening State Invariants Generated from Requirements Specifica-tions. In Proceedings of the 5th International Symposium on Requirements Engi-neering, RE’01, Toronto, Canada. IEEE.

[Jeusfeld and Johnen, 1994] Jeusfeld, M. and Johnen, U. (1994). An Executable MetaModel for Re-Engineering of Database Schemas. In Proceedings of the 13th Inter-national Conference on Entity-relationship Approach, Manchester, UK.

[Jouanot et al., 2000] Jouanot, F., Nicolle, C., Cullot, N., and Ytongnon, K. (2000).Information Source Interoperability Using Context Matching Methodology. In Pro-ceedings of the International Conference Parallel and Distributed Computing Sys-tems , PDCS’2000.

18

[Jurjens, 2001] Jurjens, J. (2001). Towards Development of Secure Systems UsingUML. In Humann H. editor, Fundamental Approaches to Software Engineering.LNCS 2029, pages 187–200.

[Kakoudakis, 1996] Kakoudakis, I. (1996). The TAU Temporal Object Model. Mphilthesis, UMIST, Manchester, UK.

[Koch et al., 2000a] Koch, C., Goff, Z. K. J. L., McClatchey, R., Petta, P., andSolomonides, T. (2000). Explicit Modeling of the Semantics of Large Multi-layeredObject-Oriented Databases. In Proceedings of the the 19th International Conferenceon Conceptual Modelling, ER’2000.

[Koch et al., 2000b] Koch, C., Petta, P., Goff, J. L., and McClatchey, R. (2000b). OnInformation Integration in Large Scientific Collaborations. Technical Report TR-2000-28, sterreichisches Forschungsinstitut fr Artificial Intelligence, Wien, Austria.

[Koch et al., 2000] Koch, N., Baumeister, H., Hennicker, R., and Mandel, L. (2000).Extending UML for Modeling Navigation and Presentation in Web Applications. InProceedings of the Workshop Modeling Web Applications in the UML, UML’00.

[Leclercq, 2000] Leclercq, E. (2000). Interoperabilite semantique des systemesd’information geographique : une approche basee sur la mediation de contexte. Thsede doctorat, Universite de Bourgogne, UFR Sciences et Techniques.

[Ledeczi et al., 2001] Ledeczi, A., Maroti, M., Karsai, G., Garrett, J., Thomason, C.,Nordstrom, G., Sprinkle, J., and Volgyesi, P. (2001). The Generic Modeling Envi-ronment. In Proceedings of the WISP’2001 Conference.

[Lou, 1997] Lou, J. (1997). Data Model Description and Translation Using the Meta-model SOME. Phd thesis, EPFL, Lausanne, Switzerland.

[Lyons, 1998] Lyons, A. (1998). UML for Real-Time Overview. Available at URLwww.objectime.com/otl/technical.

[Medvidovic and Rosenblum, 1999] Medvidovic, N. and Rosenblum, D. S. (1999).Assessing the Suitability of a Standard Design Method for Modeling Software Ar-chitectures. In Proceedings of the 1sr IFIP Working Conference on Software Archi-tecture, San Antonio, Texas, USA, pages 161–182.

[Medvidovic et al., 1996] Medvidovic, N., Taylor, R. N., and E. James Whitehead, J.(1996). Formal Modeling of Software Architectures at Multiple Levels of Abstrac-tion. In Proceedings of the California Software Symposium 1996, pages 28–40.

[Mili et al., 1995] Mili, H., Pachet, F., Benyahia, I., and Eddy, F. (1995). WorkshopSummary. In Proceedings of OOPSLA’95 Workshop on Metamodeling in OO, pages105–110, Austin, TX, USA.

[Nicolle, 1996] Nicolle, C. (1996). Traduction multi-modles dans les systmesd’information coopratifs. Doctorat d’informatique, Universite de Bourgogne, UFRSciences et Techniques.

19

[Nordstrom et al., 1999] Nordstrom, G., Sztipanovits, J., Karsai, G., and LEDECZI,A. (1999). Metamodeling-Rapid Design and Evolution of Domain-Specific Mod-eling Environments. In Proceedings of the IEEE Conference and Worshop on En-gineering of Computer-Based Systems, ECBS’99, pages 68–74. Available at URLhttp://computer.org/proceedings/ecbs/0028.

[Ober, 2000] Ober, I. (2000). Difficulties in Defining Precise Semantics for UML.ECOOP’00 Workshop 14 on Defining a Precise Semantics for UML, Sophia An-tipolis, France.

[OMG, 1999] OMG99-08-02 (1999). A UML Profile for CORBA, OMG Report 99-08-02. Available at URL http://www.omg.org, Version 1.0, August 2, 1999.

[Opdahl, 1998] Opdahl (1998). RAISA Project. Available at URLhttp://www.ifi.uib.no/projects/raisa/.

[Opdahl, 1996] Opdahl, A. L. (1996). A Model of the IS-architecture Alignment Prob-lem. In Proceedings of VITS Autumn Conference on System Architectures, Bors,Sweden.

[Pons et al., 1998] Pons, C., Waldoke, S., Paz Mezzano, C., and Felder, M. (1998). AFormal Approach to Practical Object-Oriented Analysis and Design. In Proceed-ings of Argentinean Symposium of Object Orientation ASOO’98, Buenos Aires, Ar-gentina. Available at URL http://www-lifia.info.unlp.edu.ar/vcpons.

[Price et al., 1999] Price, R., Ramamohanarao, K., and Srinivasan, B. (1999). Spa-tiotemporal Extensions to Unified Modeling Language. In Proceedings of the 10th

International Worshop on Database and Expert Systems Applications, pages 460–461. IEEE.

[Robbins et al., 1998] Robbins, J. E., Medvidovic, N., Redmiles, D. F., and Rosen-blum, D. S. (1998). Integrating Architecture Description Languages with a StandardDesign Method. In Proceedings of the 1998 International Conference on SoftwareEngineering, pages 209–218. IEEE.

[Schatz and Huber, 1999] Schatz, B. and Huber, F. (1999). Integrating Formal De-scription Techniques. In Wing, J. M., Woodcock, J., and Davies, J., editors, FM’99– Formal Methods, volume 2 of LNCS 1709, pages 1206–1225. Springer Verlag.Available at URL http://www4.informatik.tu-muenchen.de/papers.

[SCSC, 1999] SCSC (1999). Preface. In Proceedings of the Information SystemsFoundations Workshop Ontology, Semiotics and Practice, Lausanne, Switzerland.

[Selic and Rumbaugh, 1998] Selic, B. and Rumbaugh, J. (1998). UsingUML for Modeling Complex Real-Time Systems. Available at URLhttp://www.objectime.com/otl/technical/umlrt.html.

[Svinterikou and Theodoulidis, 1997] Svinterikou, M. and Theodoulidis, B. (1997).The Temporal Unified Modelling Language TUML. Technical Report TR-97-1,TimeLab, Department of Computation, UMIST, UK.

20

[Terrasse, 2001] Terrasse, M.-N. (2001). A Metamodeling Approach to Evolution. InBalsters, H., de Bruck, B., and Conrad, S., editors, Database Schema Evolution andMeta-Modeling. Springer-Verlag, LNCS 2065, ISBN 3-540-42272-2, pages 202–219. 9th International Workshop on Foundations of Models and Languages forData and Objects, Schloss Dagstuhl, Germany, September 2000.

[Terrasse and Savonnet, 2000] Terrasse, M.-N. and Savonnet, M. (2000). Formaliza-tion of the UML Metamodel: An Approach Based Upon the Four-Layer Metamod-eling Architecture. ECOOP’00 Workshop 14 on Defining a Precise Semantics forUML, Sophia Antipolis, France.

[Terrasse et al., 2001] Terrasse, M.-N., Savonnet, M., and Becker, G.(2001). An UML-metamodeling Architecture for Interoperability ofInformation Systems. In Proceedings of the International Confer-ence on Information Systems Modelling, ISM’01. Available at URLhttp://www.fee.vutbr.cz/UIVT/ism/ISM programElEdition.htm.

[UML, 2000] OMG (2000). OMG Unified Modeling Language Specification. Version1.3, Available at URL http://www.omg.org.

[Wijers, 1991] Wijers, G. (1991). Modelling Support in Information Systems Devel-opment. Phd thesis, Delft University of Technology, The Netherlands.

[Zhou et al., 2000] Zhou, C., Lu, F., and Wan, Q. (2000). A Conceptual Model for aFeature-Based Virtual Network. GeoInformatica, 4(3):271–286.

21