Date of Award

Fall 1998

Document Type

Dissertation

Degree Name

Doctor of Philosophy in Computing Sciences - (Ph.D.)

Department

Computer and Information Science

First Advisor

Wilhelm Rossak

Second Advisor

Alexander D. Stoyenko

Third Advisor

Franz J. Kurfess

Fourth Advisor

Harold Lawson

Abstract

The ever growing complexity of software systems has revealed many short-comings in existing software engineering practices and has raised interest in architecture-driven software development. A system's architecture provides a model of the system that suppresses implementation detail, allowing the architects to concentrate on the analysis and decisions that are most critical to structuring the system to satisfy its requirements. Recently, interests of researchers and practi-tioners have shifted from individual system architectures to architectures for classes of software systems which provide more general, reusable solutions to the issues of overall system organization, interoperability, and allocation of services to system components. These generic architectures, such as product line architectures and domain architectures, promote reuse and interoperability, and create a basis for cost effective construction of high-quality systems. Our focus in this dissertation is on domain architectures as a means of development and integration of large-scale, domain-specific business software systems.

Business imperatives, including flexibility, productivity, quality, and ability to adapt to changes, have fostered demands for flexible, coherent and enterprise--wide integrated business systems. The components of such systems, developed separately or purchased off the shelf, need to cohesively form an overall compu-tational environment for the business. The inevitable complexity of such integrated solutions and the highly-demanding process of their construction, management, and evolution support require new software engineering methodologies and tools. Domain architectures, prescribing the organization of software systems in a business domain, hold a promise to serve as a foundation on which such integrated business systems can be effectively constructed.

To meet the above expectations, software architectures must be properly defined, represented, and applied, which requires suitable methodologies as well as process and tool support. Despite research efforts, however, state-of-the-art methods and tools for architecture-based system development do not yet meet the practical needs of system developers.

The primary focus of this dissertation is on developing methods and tools to support domain architecture engineering and on leveraging architectures to achieve improved system development and integration in presence of increased complexity. In particular, the thesis explores issues related to the following three aspects of software technology: system complexity and software architectures as tools to alleviate complexity; domain architectures as frameworks for construction of large scale, flexible, enterprise-wide software systems; and architectural models and representation techniques as a basis for "good” design. The thesis presents an archi-tectural taxonomy to help categorize and better understand architectural efforts. Furthermore, it clarifies the purpose of domain architectures and characterizes them in detail.

To support the definition and application of domain architectures we have developed a method for domain architecture engineering and representation: GARM-ASPECT. GARM, the Generic Architecture Reference Model, underlying the method, is a system of modeling abstractions, relations and recommendations for building representations of reference software architectures. The model's focus on reference and domain architectures determines its main distinguishing features: multiple views of architectural elements, a separate rule system to express constraints on architecture element types, and annotations such as “libraries” of patterns and “logs” of guidelines. ASPECT is an architecture description language based on GARM. It provides a normalized vocabulary for representing the skeleton of an architecture, its structural view, and establishes a framework for capturing archi-tectural constraints. It also allows extensions of the structural view with auxiliary information, such as behavior or quality specifications. In this respect, ASPECT provides facilities for establishing relationships among different specifications and gluing them together within an overall architectural description. This design allows flexibility and adaptability of the methodology to the specifics of a domain or a family of systems. ASPECT supports the representation of reference architectures as well as individual system architectures. The practical applicability of this method has been tested through a case study in an industrial setting.

The approach to architecture engineering and representation, presented in this dissertation, is pragmatic and oriented towards software practitioners. GARM-ASPECT, as well as the taxonomy of architectures are of use to architects, system planners and system engineers. Beyond these practical contributions, this thesis also creates a more solid basis for expbring the applicability of architectural abstractions, the practicality of representation approaches, and the changes required to the devel-opment process in order to achieve the benefits from an architecture-driven software technology.

Share

COinS