Abstract: Context-aware applications are subject to inevitable context inconsistencies due to various reasons including unpredictable and uncontrollable sensing noises in pervasive computing environment. These inconsistencies can lead to applications abnormal or even failure. Various approaches have been proposed to automatically resolve the detected context inconsistencies for such applications. However, it is not clear whether maintaining consistency for contexts at a data level can indeed improve an application’s quality at a software level. This paper designs a controlled experimentation to empirically study this problem, based on a group of selected context-aware robot car applications. Then, this paper compares and reports the benefits and side effects of different context inconsistency resolution approaches. The experimental results can help developers understand how context inconsistency and its resolution impact context-aware applications, as well as improving the dependability for these applications.

Key words: pervasive computing, context inconsistency, controlled experimentation

摘要： 在普适计算环境中，上下文信息由于受到噪声等不可控因素的干扰，面临着一致性错误的问题。这些错误会影响应用的正常运行，使其表现异常甚至失效。目前已经有自动的一致性错误处理方法来保障应用所获取的上下文满足数据层面的一致性约束，然而在数据层面保持一致是否能提升应用的运行质量并不明确。系统地研究了这个问题，并基于一组真实的上下文感知机器人小车应用，设计了一系列受控实验，来分析和比较在模拟和真实世界中上下文一致性错误的处理方式对应用运行质量的改善程度及潜在的负面影响。利用该实验的分析结果，将有助于提高普适计算中上下文感知应用的运行质量。

关键词: 普适计算, 上下文一致性错误, 受控实验