Model-Based Runtime Adaptation of Resource Constrained Devices

Dynamic Software Product Line (DSPL) engineering represents a promising approach for planning and applying runtime reconfiguration scenarios to self-adaptive software systems. Reconfigurations at runtime allow those systems to continuously adapt themselves to ever changing contextual requirements. With a systematic engineering approach such as DSPLs, a self-adaptive software system becomes more reliable and predictable. However, applying DSPLs in the vital domain of highly context-aware systems, e.g., mobile devices such as smartphones or tablets, is obstructed by the inherently limited resources. Therefore, mobile devices are not capable to handle large, constrained (re-)configuration spaces of complex self-adaptive software systems. The reconfiguration behavior of a DSPL is specified via so called feature models. However, the derivation of a reconfiguration based on a feature model (i) induces computational costs and (ii) utilizes the available memory. To tackle these drawbacks, I propose a model-based approach for designing DSPLs in a way that allows for a trade-off between pre-computation of reconfiguration scenarios at development time and on-demand evolution at runtime. In this regard, I intend to shift computational complexity from runtime to development time. Therefore, I propose the following three techniques for (1) enriching feature models with context information to reason about potential contextual changes, (2) reducing a DSPL specification w.r.t. the individual characteristics of a mobile device, and (3) specifying a context-aware reconfiguration process on the basis of a scalable transition system incorporating state space abstractions and incremental refinements at runtime. In addition to these optimization steps executed prior to runtime, I introduce a concept for (4) reducing the operational costs utilized by a reconfiguration at runtime on a long-term basis w.r.t. the DSPL transition system deployed on the device. To realize this concept, the DSPL transition system is enriched with non-functional properties, e.g., costs of a reconfiguration, and behavioral properties, e.g., the probability of a change within the contextual situation of a device. This provides the possibility to determine reconfigurations with minimum costs w.r.t. estimated long-term changes in the context of a device. The concepts and techniques contributed in this thesis are illustrated by means of a mobile device case study. Further, implementation strategies are presented and evaluated considering different trade-off metrics to provide detailed insights into benefits and drawbacks