Towards Dependable Fixing Of Context Inconsistency In Perva- Sive Computing

In pervasive computing, environmental contexts are subject to frequent changes. Context-aware applications need to adapt their state and behavior accordingly. Howev-er, context inconsistency can occur easily due to many reasons including unpredicted environmental noises and dynamics. Context inconsistency may cause context-aware applications to behave abnormally or even fail. This calls for the need of automated fixing of detected context inconsistencies. However, some of existing techniques had strong assumptions that may not hold in practice, some proposed to remove suspicious contexts based on various heuristics without guaranteeing the soundness of doing so, and others required human participation, deviating from our goal of automated fix-ing for context inconsistency. Therefore, existing work does not address this context inconsistency fixing problem satisfactorily.In this thesis, we proposed three novel techniques aiming at automatically fixing context inconsistency. These techniques are based on a set of formalizations and logical inductions. They are namely complete-fixing, conservative-fixing and hybrid-fixing. The three techniques generally work in three stages:(1) detecting context inconsisten-cy; (2) generating abstract repair cases, and (3) executing them concretely to validate whether context inconsistency has been fixed. The complete-fixing technique can en-sure the completeness of fixing, but it is at the cost of reduced time and space efficiency. The conservative-fixing technique can ensure the soundness of fixing by locking shared elements when analyzing their dependencies inside consistency constraints. It works more efficiently than the complete-fixing technique, but may not necessarily fix con-text inconsistency at certain scenarios. Finally, the hybrid-fixing technique combines static analysis of consistency constraints and dynamic generation of repair actions to ensure the soundness of its generated repair cases, even when element dependencies exist inside consistency constraints. It can significantly increase the success rate for fixing context inconsistency, and at the same time incur only very small time cost.We evaluated our three techniques experimentally. The experimental results re-ported that:First, our techniques are useful for addressing practical context inconsis-tency problems; Second, our techniques have satisfactory scalability, which enables them to handle large-sized context inconsistency problems with small cost; Third, our techniques are flexible and can be easily integrated with efficient context inconsistency detection techniques, e.g., partial constraint checking technique, to address practical context inconsistency problems. In summary, our techniques have distinct advantages over existing techniques on fixing context inconsistency, and can provide automation, soundness and efficiency.