| This dissertation addresses the problem of software monitoring in mobile systems. Traditional software monitoring simply collects raw data from running applications in real time and performs postmortem data analysis. This paradigm is unsuitable for mobile mon itoring due to resource constraints and disconnected operation in mobile platforms. Mobile Surveyor, a model-based monitoring system presented in this dissertation, provides a new perspective and approach to mobile monitoring.; The overall research contribution of Mobile Surveyor is the incorporation of data models throughout the monitoring process. The idea of model-based monitoring is to collect data items that deviate from a predictive model during monitoring and restore data values based on a descriptive model at the postmortem data analysis stage. In this way, Mobile Surveyor addresses the challenges of limited resources in mobile monitoring by achieving less memory and bandwidth usage in data collection and transmission.; Building accurate and appropriate models is the core task in Mobile Surveyor. One contribution of this dissertation is to build a predictive model that is composed of a monitoring strategy and an adaptation algorithm. This approach distinguishes Mobile Surveyor from traditional monitoring tools by dynamically making monitoring decisions in real time based on predictive models. Another contribution is the adaptation of a model search and optimization algorithm from data mining methods to iteratively build predictive models. We use curve fitting methods to construct the descriptive models. Two types of agents, mobile agents and inference agents, are implemented. Mobile Surveyor utilizes the autonomy of mobile agents to perform mobile monitoring by constructing agents based on predictive models. Inference agents are responsible for data analysis based on descriptive models.; Mobile Surveyor turns the user from a passive receiver of information to a proactive one that participates in monitoring. It captures a user's monitoring requirements as a set of quality of service (QoS) specifications and embeds them in mobile agents. This provides another way to save monitoring resources by only collecting what users need.; A Palm Pilot Robot is used as the testbed for a case study of model-based monitoring in mobile systems. The experimental results demonstrate the success of Mobile Surveyor. |
