Autonomic Business-Driven Decision-Making for Adaptation of Service-Oriented Systems

摘要

When business or technical changes occur in business processes during runtime, the affected service-oriented systems, including Web service compositions, should be adapted to accommodate the changes. This adaptation usually can be done in several ways. Deciding the most suitable adaptation way is very critical and complicated. Therefore, advanced decision-making is needed to determine how to proceed the adaptation. The majority of past service-oriented system management work focused on optimization of quality of service but such optimisation might not lead to maximisation of business value. The aim of this research is to propose optimisation models, decision-making algorithms, policy language constructs and architecture components for building software that makes runtime instance-based decisions for adaptation of service-oriented systems in ways that maximise overall business value, while satisfying all given constraints, such as business value constraints, cost constraints, time constraints, and resource constraints.The research contributions include: Classifications of problems in decision-making for adaptation of Web service compositions; A set of problem models in constraint programming for several common situations that require runtime adaptation decision-making in service-oriented systems; A set of business-driven instance-based adaptation decision-making algorithms enacting the developed constraint programming models, as well as discussion how different adaptation situations influence design of these algorithms; A set of policy language constructs supporting adaptation decision-making; Architecture of software for runtime adaptation decision-making that optimizes business value in the examined common situations. The proposed solutions have been validated and evaluated through prototype implementation, case studies and experiments in terms of feasibility, functional correctness, business benefits, performance and scalability. The results of these evaluations showed that 1) it is feasible to use the proposed algorithms, middleware architecture and policy constructs to make adaptation decisions; 2) The proposed algorithms are functional correct; 3) the underlying optimisation models are adequate; 4) The decisions made by the proposed business-driven instance-based decision-making algorithms can lead to better business benefits compared to single adaptation decision-making algorithms; 5) The proposed algorithms are scalable. It is crucial to note that these solutions can be adapted to various software systems (the used policy language and the software middleware are only examples).