Achieving System-Wide Architectural Qualities

System-wide properties such as reliability, availability, maintainability, security, responsiveness, adaptivity, evolvability, survivability, nomadicity, manageability, and scalability (the \ilities"), are crucial for the success of large software systems. Although these properties have been a major concern of software engineering since its inception, most of the eeort on software architecture has focused on achieving functionality. For example, in current visions of component software architectures (CORBA, WWW, ActiveX, etc.), there is no provision for systematically achieving system-wide properties. As noted in the objectives statement of this workshop WCSA98], \assembling components and also achieving system-wide qualities is still an unsolved problem. As long as the code that implements ilities has to be tightly interwoven with code that supports business logic, new applications are destined to rapidly become as diicult to maintain as legacy code." Given a requirements description as the problem statement, there can potentially be a combinato-rially large number of architectural design alternatives as solutions. For example, an architectural design involves deciding on the number and types of components in the system, the number and types of interactions, the way data is distributed among components, the way processing is distributed among components, and so on. Inevitably decisions have to be made on these choices toward a particular nal system architecture. Clearly, the quality of the architecture chosen is only as good as the decisions taken to arrive at it. What then is the relationship between system-wide qualities and architectural design? If we view architectural design in terms of a design space, then the ilities are constraints on that space. If we view the process of architectural design as the incremental construction of the design space and the progressive narrowing down of that space towards a \good enough" solution, then the ilities or system-wide qualities can be held as goals to explicitly guide the generation of alternatives at each step, and to guide the selection among alternatives throughout the process. When building systems from components, one needs to pay attention to many system-wide qualities at the same time, since they can interact with each other in many ways. These qualities and their interactions must be identiied and analyzed for each design decision. For example, in the absence of any general theories that can relate reliability, maintainability, scalability and a