Inconsistency detection and resolution for context-aware pervasive computing

Pervasive computing environments are often noisy and subject to change. Although software should be responsive to contexts by changing their behaviors, the contexts themselves may be abnormal or imprecise. This results in context inconsistencies, that is, the conflicts among contexts. Context inconsistencies may set such software into a wrong state or lead software to wrongly adjust their behaviors. It is desirable to detect and then resolve these inconsistencies in time to prevent software from misbehaving. One popular approach is to detect inconsistencies when contexts breach certain consistency constraints. Existing constraint checking techniques recheck the entire expression of each affected constraint upon context changes. However, when a changed context affects only a constraint's sub-expression, rechecking the entire expression introduces delays to detecting other inconsistencies. This thesis proposes a formal model and its supporting algorithms that identify the parts of previous checking results that are reusable without missing context inconsistencies identifiable via entire rechecking. This enables us to combine the reusable and rechecked parts to produce the final results efficiently. Our evaluation reports a more than fifteenfold performance improvement with our approach against conventional approaches for detecting context inconsistencies. An important, follow-up step is to resolve detected inconsistent contexts automatically for applications. However, the effectiveness of existing resolution strategies is compromised by their formulated assumptions that do not fully hold in practice. To different extents, this makes applications using the resolved contexts less context-aware than the ideal case. This thesis proposes the drop-bad and the impact-oriented strategies. With the same goal of protecting context-awareness, the two strategies are formulated on an intuitive observation and the applications' situation specifications, respectively. Our evaluation reports that both strategies protected at least 15% more situations for context-aware applications than existing resolution strategies without affecting context inconsistency resolution