Context-aware Mobile Computing in the Cloud: Context-driven Decision-making for Self-Adaptation

Decision optimization for computation offloading is a novel research area which is gaining popularity due to the proliferation of advanced sensor-based mobile applications and the growing compute and data resources in advanced mobile devices. The work presented in this thesis fits into the area of context-aware mobile application development in distributed cloud environment with a focus ondynamicdecisionsupport. Thisthesishighlightstheneedofanopportunistic dynamic decision-making towards the context-driven configuration and selfadaptation for computation offloading tackling the challenges of the varying resource performance trade-offs, the influence of the runtime uncertainty in the context and the conflicting offloading objectives. The presented decision support can ascertain the quality of its own decision applying directed probabilistic graphical models at runtime. Computation offloading is an interesting research field driven by the advanced connectivity and the resource-constrained devices. In literature, the focus of computation offloading revolves around how, what, where and when to offload the data or processing. The how and what questions grabbed the attention of researchers most and it is a saturated area exploring different techniques of code partitioning and state transitions. We have focused more the rest of the two focus areas due to the fast pace of technological advancements in mobile technology. Mobile devices are more powerful than ever. We have investigated their threshold to use the cloud resources for data-intensive use-cases and presented many resource-performance trade-offs. An enhanced benchmarking framework is also contributed to benchmark the performance, memory scalability and energy consumption for context-aware applications. Context-aware configurations and adaptations enable the smartness and an ubiquitous interaction with the users characterized by the subjectivity and the dynamics owing to the mobility and available resources on the device. The key feature of smart mobile applications is that the data is generated on the mobile devices. The subjective context and QoS requirements regarding the type of the application, the level of the smartness, i.e. the frequency of the adaptations or the type of the context data, the device offerings, and the context-driven configuration of the sensors introduce resource performance trade-offs on the fly. Several dynamic decision models are designed to tackle the inherent uncertainty, subjectivity, and dynamism analysing the role of QoC in decision-making for context-aware self-adaptations. We further investigate and provide decision support under QoE requirements for smoother interaction maintaining the QoS in terms of lower latency under a Qo*-aware fashion. MAsCOT, an opportunistic decision support framework is presented for the opportunistic dynamic offloading of loosely-coupled contextaware mobile applications. A prototypical approach is used to investigate and evaluate our approach for cloud-enabled smart mobile applications. The techniques and results presented in this thesis are at the forefronts of the self-adaptive dynamic decision support. The applicability and results of the performance overhead of directed PGMs with a retrospective and reflective behaviour encourage advanced and sophisticated mobile solutions for dynamic decision-making such as distributed or hybrid decision support.status: publishe