Automated Suppression Of Context Inconsistency Hazards With Pattern Learning

Context-aware applications in pervasive computing environment can automatically adapt according to context information collected from sensors to provide smart services for users. However, due to imperfect nature of sensing devices and data transmission, contexts can be inaccurate, which leads to the context inconsistency problem. Context inconsistencies need to be detected and resolved in time. Otherwise applications might fail to provide services normally, or even crash. Contexts are usually managed by mid-dleware systems, which can deploy a set of consistency constraint to specify properties that must hold concerning contexts. These constraints are evaluated when contexts are changed and if any constraint is violated, an inconsistency is detected. This process is called the context inconsistency detection and resolution.Traditional strategy is to evaluate related constraints whenever any context is changed, and resolve inconsistencies immediately after detection. However, this s-trategy is subject to numerous false alarms. A large part of detected inconsistencies disappear spontaneously after a small set of later context changes are applied, and do no harm to applications. Such false alarms are named as inconsistency hazards, and hazards for short. Detection and resolution for inconsistency hazards are unnecessary, and might even be harmful. Thus we need to suppress such false alarms.However, inconsistency hazards resemble normal inconsistencies since they are both caused by violation of consistency constraints. The only difference between them is that hazards disappear quickly after emerge. This article analysis the hazard problem and proposed a novel approach to schedule inconsistency detection automatically for hazard suppression. The key observation is that inconsistency detection is subject to such inconsistency hazards only under certain patterns of context changes. With this knowledge, inconsistency detection can be reasonably re-scheduled to suppress false alarms. Effectiveness of the approach proposed in this article is validated through experiments, which show that more than 90% of inconsistency hazards are suppressed while most normal inconsistencies can be detected and resolved timely.