Project Objectives

The consortium enjoys a noteworthy collaboration history. Since about 20 years Joachim Schmidt and John Mylopoulos have cooperated closely in several research projects. In recent years this cooperation extended to Dr. Matthias Jarke, Technical University of Aachen and to Dr. Carson Woo at the University of British Columbia (see also section 3).

In the late 80ies Drs. Schmidt , Mylopoulos, Jarke and other European partners cooperated successfully for about five years in the ESPRIT project DAIDA on design support for data-intensive applications. The Canadians provided formal models for requirement specification and theoretical results for mapping specifications into designs. The European side contributed its achievements in advanced implementation technology such as database programming languages and other tools for mapping information system designs into implementations. In retrospect, DAIDA is considered as one of the first successful projects in object-oriented information system design and maintenance (in terms of technology development and transfer, publications, advanced teaching tools etc.).

For the cooperation which is now under consideration the mutual interests are again defined by a similar pattern of complementary expertise. Our prospective EU/Canadian collaboration extends DAIDA's more static approach to program and database development into the dynamic dimension of long-term, cooperating processes and open, heterogeneous systems. Since 1990 the major DB/PL-related European activities have been concentrated in the ESPRIT basic research project FIDE (Fully Integrated Data Environments) which is working on extended object-oriented database models and polymorphic persistent programming languages enriched by the technology that can cope with the dynamicity, longevity and heterogeneity of long-lived, data-intensive activities.

FIDE and related projects are now in a similar position as DAIDA was in the late 80ies: they hold major parts of the technology for object- and activity-oriented information systems in heterogeneous environments. However, to support application development using this technology there is a strong need for relating the entire range of experience including adequate formal foundation for conceptual modelling, functional and non-functional requirement specification, script languages, organizational computing, workflow experience etc. FIDE and related projects can provide the enabling system technology which can handle adequately issues of extensible data modelling and typing, type-safe polymorphic data manipulation, higher-order functionality for program generation and transformation, persistence and distribution, commercial server integration, etc.

Since approximately 15 years John Mylopoulos, Joachim Schmidt and Matthias Jarke have together organized a series of workshops on communicating intelligent data-intensive systems. Several PhD theses on the subject have been jointly supervised (e.g. Chung, Wetzel, Schael). Further collaboration preparation on Cooperative Information Systems was done in a cooperative effort between Eric Dubois and John Mylopoulos. Between Aachen, Toronto, and Hamburg there has been exchanges of research personnel as well as between Milano and Aachen.

With the announcement last year of the effort between the European Union and Canada in establishing a research cooperation programme, the consortium initiated intensive discussions on cooperation. In April 1994 Mr. Brian Nixon of the Canadian side visited Hamburg for an exchange on cooperation topics and familiarization of systems and tools available. Hamburg University demonstrated its high interest in the prospected cooperation by financing all of Mr. Nixon's costs here at Hamburg.

2.4 Complementary Cooperation Experiences in Research and Development

While the European partners are particularly experienced in Computational Modelling with a strong background in Systems Technology, the Canadian side has complementary expertise in Conceptual Modelling with a strong background in Artificial Intelligence. Both sides have a solid formal background and a strong record in computer application and system implementation as well as in project cooperation:

European side:

Giorgio De Michelis: At the Milan side an initial conceptual framework for conceptualizing the requirements of mechanisms of interaction in a systematic way as well as for design and evaluation of mechansims of interaction has been developed. This framework distinguishes two complementary constitutive aspects of a mechanism of interaction:

• Notations for controlling and specifying the behaviour of interaction mechanism;
• Facilities for accessing and manipulating interaction mechanism;
• Exception handling mechanisms for Workflow Management Systems;
• Workflow Execution Modules and change management.

The Milan Workflow Model is based on Petri Nets, as is the case in many previous Workflow Management Systems. Petri Nets are widely used for specifying many sorts of mechanisms of interaction within Computer Supports for Cooperative Work. But on the one hand it is based on a small subclass of Elementary Net Systems with simple elements and highly readable models, and on the other it is based on some nice mathematical properties of that class, allowing the generation of a large class of behaviours from those models. Moreover, Milan is able to demonstrate that some interesting features proposed in the literature, for example by Ellis & Keddara, 1993, to enhance the usability of Workflow Management Systems can be easily embedded within it.

Most of the languages supporting these different activities have been proven useful for the expression of requirements related to classical centralized "business" information systems. However, major deficiencies have been pointed out with respect to their capabilities for expressing new kinds of requirements like those related to real-time performances, complex data structures, security aspects and concurrency/cooperation protocols in distributed Cooperative Information Systems. In the light of recent advances in organizational modelling and distributed AI, a consensus is now being reached in favour of the development of a new generation of languages for capturing requirements inherent to Cooperative Information Systems.

Namur has investigated the use of ALBERT at the modelling and analysis levels. More recently, using the language in the context of large manufacturing Cooperative Information Systems, Namur has found that the validation activity is also crucial. Besides the use of a graphical formalism with "boxes and arrows" with which the customer is familiar, the need for a better support has been identified providing help to the customer for checking the adequacy of the behaviour displayed by the system formally specified.

Consequently, one can characterize the degree of reuse in terms of a channel of communication between the original developers and the re-users. The broader and better defined the channel, the greater the potential for reuse, and therefore for productivity improvements.

• For program libraries, for example, the channel is well-defined but narrow since all development experience other than coding is missing.
• For experience-sharing meetings between original developer and re-user the channel is broad but ill-defined as it relies on human memory.

At the Aachen side the insight gained in a series of CASE integration projects is exploited by defining a data modelling language for respository managers and their abstractional, assertional and dynamic clustering requirements. Aachen and Toronto have gained extensive experience which illustrates how the deductive object management system ConceptBase embodying the knowledge representation language TELOS as its data model meet these requirements.

The Aachen side has developed an approach which allows the support of traceability in information systems. This approach leads to the storage of all information which appears during the running of the development process. This process information allows an efficient and comprehensible adaptation of information systems to changing relationships or a changed environment. The acceptance of the system by developers and users is also increased through understanding the system evolution. Both aspects - process trace und process guidance - were researched separately primarily on account of their contrasting differences. Research has shown, however, that both approaches can increase greatly the productivity of development in information systems.

Tycoon provides essential supports for the following two activities in database application development:

• Generic server integration: By virtue of Tycoon's polymorphic (higher-order) type system it is possible to also integrate pre-existing, independently developed generic servers (like object-oriented databases, C++ GUI libraries or RPC communication services) as strongly typed parametric libraries into the Tycoon programming environment. Therefore, systems developed in Tycoon fit smoothly into open system architectures.
• High-level application engineering: Tycoon is used by application programmers to realize the full functionality of data-intensive applications which require a semantically close interaction between objects on the screen, objects in main memory, objects on disk, and objects on the wire. Tycoon supports such tasks by providing uniform and generalized naming, typing and binding concepts that abstract from the specifics of the underlying object servers like GUI toolkits, programming languages, database systems and RPC services.

The open, library-based approach to system construction is currently being pursued in several system frameworks that are based on C++ or distributed object models of similar expressiveness. Tycoon aims at a higher system development productivity in a language framework with the following characteristics:

• Improved language orthogonality: All language entities in Tycoon (like values, functions, modules and types) have first class status. For example, it is possible to write a Tycoon function that receives a type as its argument and returns a module which aggregates a set of dynamically constructed functions for a fresh abstract data type. Such higher-order language concepts are particularly helpful to factor-out complex but repetitive programming tasks from individual applications into shared, reusable library code.
• Increased type system expressiveness: Tycoon combines subtype and parametric polymorphism. Furthermore, both forms of polymorphism are generalized to (higher-order) type operators supporting the type-safe definition of highly polymorphic system libraries.
• Orthogonal persistence abstraction:Tycoon developers do not have to distinguish between local volatile data and shared global and persistent data. As a consequence, programmers can fully abstract from store properties (size of main memory, garbage collection, transfer between primary and secondary store, data format conversion between nodes in heterogeneous networks, etc.).
• Reflective programming support: Some system tasks in data-intensive applications (e.g., query optimization, transaction scheduling, GUI generation) are based on run-time reflective programming techniques. Run-time linguistic reflection denotes the ability of a system to inspect parts of the system (e.g. query expressions, transaction instruction sequences, type structures) at run-time and to dynamically extend or modify the system based on the outcome of this inspection.
• Persistent threads: The Tycoon system provides persistent threads as a novel form of bindings from data in persistent object stores to activated code. This form of binding generalizes existing data and code binding concepts and is particularly well-suited for applications that manage long-term, distributed or cooperative activities. Persistent threads, exhibit an interesting duality. On the one hand side threads can be viewed as activities that can be created, executed, synchronized, suspended, terminated, etc. Alternatively, they can be viewed as passive data that can be stored persistently, annotated with attributes, associated with other persistent data structures, moved between nodes in a network, and manipulated by computations. This duality makes persistent threads a key technology for future persistent object systems and for applications that manage long-term, distributed or cooperative activities like computer-aided design, workgroup communication and workflow management.

Canadian side:

John Mylopoulos: The research group in Toronto is working on a framework for representing and using Non-Functional Requirements (NFRs) for system development. Results of this research fall into four categories:

• a process-oriented NFR framework
• applications to several classes of NFRs
• prototype implementation of the framework, and
• an assessment of the extent to which the framework is useful in realistic settings.

The Toronto framework allows NFRs to be represented as (potentially) harmonious or conflicting goals which can be decomposed, satisfied, and supported by the use of various generic methods. Correlation rules detect interactions induced by the use of generic methods, warn the designer and prevent certain actions that might jeopardize the satisfying of requirements. The framework captures design rationale, makes trade-offs explicit in a goal graph structure, and evaluates their effects through the labelling procedure. Throughout the process NFRs serve as selection criteria for choosing among a myriad of decisions which then act as a basis for justifying the overall design.

Applications to several classes of NFRs, with primary emphasis on accuracy and security requirements, have shown that there is the need for capturing various types of available development knowledge specific to such NFR goals, and that the framework meets the need in terms of generic methods and correlation rules.

Applications to several classes of NFRs, with primary emphasis on accuracy and security requirements, have shown that there is the need for capturing various types of available development knowledge specific to such NFR goals, and that the framework meets the need in terms of generic methods and correlation rules.

Through the implementation of the NFR assistant, it was found that the NFR framework is amenable to automating parts of the development process. The portions that are automated include mundane bookkeeping as well as more complex generation of offspring via method hierarchies or correlation tables as well as generation of correlation links and realization of the labelling procedure. By applying its NFR framework to realistic settings Toronto has shown three important benefits of the framework:

• It has been demonstrated that there is a strong need for systematic representation of NFRs and that the NFR framework meets the need. Some important observations made were as follows. Different application domains have different classes of NFRs. Goal descriptions in the source documents are often ambiguous, but differentiating goals into NFR, satisfying and argument goals helps deal with the problem. It was observed that a good selection of initial set of NFR goals facilitates systematic rather than ad hoc development. It was also found that decomposition methods are needed in addressing the right issues for the initial set of NFR goals, and in bridging the gap between an abstract NFR goal and a specific satisfying goal. The extent to which argumentation methods are applied depends on the criticality of supporting or objecting arguments. Goal graph structures serve as a well-organized design history, showing how different parts of a design are related to one another and to the mapping of functional requriements, as well as facilitating later evolution.
• The use of the NFR assistant in applying the NFR framework to portions of a credit card system shows how automation of parts of the development process can help the developer. By browsing method hierarchies for decomposition, satisfying and argumentation, important issues in dealing with NFRs became noticeable as well as the presence of alternative methods to deal with such issues. Hence, methods offer appropriate vocabulary and subject matter for dealing with NFRs. The coverage ("hit ratio") of our collection of satisfying methods was found to be high on all three applications. Using more specialized ones requires more expert knowledge for some satisfying methods. Method taxonomies are useful firstly for capturing new methods by specializations and seondly for gradually applying more specialized methods during goal graph expansion. Methods and their taxonomies allow the designer not to explicitly represent all alternatives but important ones only.
• The NFR framework is able to help analysis of NFRs by facilitating the disambiguation of goals, the identification of the role of criticality and the detection of omissions, as well as discovering goal conflicts or inconsistencies, or goal synergies which enable avoiding redundancies and duplication of efforts.

An Organizational Computing System is a computerized system that supports organizational activities involving processing and use of data and knowledge. By "support" we mean organizational activities should be automated to the highest extent possible and the system should provide as much assistance as possible to non-automated activities. For example, finding a meeting time that does not conflict with the personal schedules of several individuals is a time-intensive task. Such a task is a good candidate for automation. However, it would be undesirable and very difficult to totally automate the meeting because some situations are unknown in advance and humans want to be able to control decision making. In this case the computer provides assistance such as tabulating voting results. The idea here is to shift as much work (i.e., work that computers perform better than human beings) as possible away from humans and onto computers.

During the past four years, Vancouver developed a basic Organizational Computing System called OASIS (Organizational Computing and Autonomous Agent Interaction). OASIS is a computer-based information system that supports the construction and deployment of business applications. It is aimed at supporting the following four major organizational needs:

• using data processing, reasoning, and communication in an integrative manner,
• allowing control over one's own applications,
• tying together independently developed applications for organizational use, and
• simplifying certain types of changes and evolution steps.

OASIS accomplishes these needs by viewing organizations as consisting of autonomous units with the capability of interacting with each other, and by the integrative use of blackboards and forms.