Previous live blog post: Generating Obstacle Conditions for Requirements Completeness
My last live blog post describes the presentation of the paper Towards Flexible Evolution of Dynamically Adaptive Systems, by Gilles Perrouin, Brice Morin, Franck Chauvel, Franck Fleurey, Jacques Klein, Yves Le Traon, Olivier Barais and Jean-Marc Jezequel (University of Namur, Belgium; SINTEF, Norway; University of Luxembourg, Luxembourg; IRISA, France). The paper was accepted at the New Ideas and Emerging Results (NIER) track, which gives only 10 minutes for presentations.
Gilles performed the presentation and started by saying that in a society of digital systems, we interact with multiple devices in a single, open, unpredictable environment. In this context, we recourse to Dynamically Adaptive Systems (DAS). In particular, their work uses an architecture-centric, model-based perspective. In the typical feedback (MAPE) loop that operationalizes the adaptation, there are many reasoning techniques (ECA rules, goal reasoning, etc.). Here, there are always trade-offs between performance and environmental uncertainty.
The problem is that the MAPE loop is usually not explicitly modeled and in many occasions the loop itself has to evolve at runtime. The proposal, therefore, is to use feature models to support the decision process and define a meta-MAPE loop that adapts the initial MAPE loop which, in turn, adapts the managed system. The meta-MAPE feature model shows different options for reasoning, configuration, runtime checking, among other MAPE activities.
Perrouin closes the presentation with two big quetions: how to engineer the meta-loop (goal-based, ECA, learning)? How would it be possible to verify (quality assurance) the meta-loops?