Beyond ATAM: Early architecture evaluation method for large-scale distributed systems

The development of large-scale distributed software systems involves substantial investment and is exposed to a high level of risk. Early architectural decisions define how the system is organised in terms of permanent data management, data communication, data input and output, coarse-grained modularisation and allocation within the organisational structure. Such a system's “back-bone” has been referred to as the System Organisation Pattern. Analysing architecture early in the development life cycle can help identify significant technical risks and mitigate them at a minimal cost. However, architecture assessment methods, such as the Architecture Trade-off Analysis Method, cannot easily be applied very early for architecture defined only conceptually. In addition, the influence of the System Organisation Pattern on the detailed properties of the final system cannot be precisely quantified, which makes applying known architecture analysis methods even more difficult. The Early Architecture Evaluation Method has been developed to assess the System Organisation Pattern much earlier than an ATAM-based assessment would be possible, i.e. in the inception phase of the Rational Unified Process. The method defines an architecture evaluation process, at the heart of which is an assessment model based on the Goal-Question-Metric scheme. The method identifies substantial risks posed by the architectural decisions comprising the System Organisation Pattern. The method has been evaluated on seven real-life examples of large-scale systems.

► Early Architecture Assessment Method is built upon GQM framework and aimed at analysis of system's buildability as well as achievability of performance and availability requirements.
► It can be applied very early in systems development lifecycle i.e. before full architecture documentation is available.
► The method identifies substantial risks posed by the early architectural decisions comprising backbone of large scale system called system organisation pattern.