A Rationale-based Model for Architecture Design Reasoning

10.5. Summary
design object may cause a change in a requirement, and the impact on the requirement
may have an impact on other disparate design elements. In this instance, forward and
backward tracings may not be sufficient to uncover all change impacts. On the other hand,
BBN can compute the likelihood of change that ripple through the network. Therefore,
these two approaches are complementary to each other in supporting design reasoning and
system maintenance activities.
It is obvious that not all system designs need to apply BBN to AREL, smaller systems
that are easy to trace and understand will gain very little from it. We suggest that large,
complex and long life-span systems that contain intricate design decisions to cater for
extensive non-functional requirements are candidates to use AREL with BBN.
The assignment of probabilities to AREL nodes individually can be laborious, it might
be a cost-saving measure to automatically assign CPTs. To make this possible, architecture
design patterns with pre-assigned probabilities are required. Architecture design patterns
can represent common ways in which decisions are made to deal with certain aspects of
a design. An example would be a security design in a system. Such architecture design
patterns would capture the essence of a design and the probabilities of change impacts.
When architecture design is carried out using design patterns, its repeated use would help
improve the accuracy of the probability estimation for change impact analysis.
10.5 Summary
In this chapter, we enhance the AREL model to provide a quantifiable reasoning structure
using the Bayesian Belief Networks. It enables architects to quantify change impacts to
architecture decisions and elements using probabilities. We have identified three different
reasoning methods in which change impacts can be analysed: (a) predictive reasoning;
(b) diagnostic reasoning; and (c) combined reasoning. With these methods, architects
can carry out quantitative analysis to predict the probability of change in an architecture
design. These methods are complementary to the traceability and qualitative analysis
methods presented in earlier chapters.
190