|Home \ Graduation Activities \ Post-Graduation Page||Login|
Integration of Strategy-oriented and Goal-oriented Approaches to Self-Adaptation
Our modern way of life is supported by complex software systems that are required to operate under substantial uncertainty, these critical systems are expected to deal with aspects such as changes in system resources, user needs, security attacks, hardware failures, etc. These critical systems usually offer adaptations that can modify and correct their behaviour, however the costs of manually maintaining them can be very high. Over the past few years, self-adaptation has established itself as an effective approach in dealing with changes in the environment in which many applications operate. Self-adaptive systems are able to adjust their behaviour at run-time in response to their perception of the environment and the system itself. There are several approaches to self-adaptation that follow the principle of separation of concerns, where the adaptation logic is kept apart from the application logic. This principle helps reduce the complexity and cost of implementing self-adaptation behaviour on the target systems. One of these approaches depends on a human operator (administrator) to supply the adaptation behaviour through adaptation conditions and strategies. This approach takes advantage of the expertise of the human operator in regards to what a problematic situation is and how best to adapt to it. However, the approach is subject to human error, and in cases where the space of possible configurations is very large finding the right adaptation strategy can be very difficult. Also, this approach can only react to problems that have been predicted by the human operator. Another approach decentralizes the knowledge over the available adaptations and their results on the system by using information provided by the developers of the various adaptable components. The human operator that manages the system uses high-level policies to configure the desired behaviour for the system. A closed-loop control layer augmented with sensors and effectors uses all this information to compute adaptation strategies at run time. This allows the approach to react to unpredictable problems. However, the computation of an adaptation plan at run time is less efficient the bigger the space of possible configurations gets and can have a considerable impact on the system. The objective of this project is to merge these two approaches in order to achieve a self-adaptation framework that can take advantage of human expertise while still being able to adapt to unpredictable situations.
Post-Graduation Student / Researcher / Professor: