A NONINTRUSIVE PACKET-BASED METHODOLOGY TO DYNAMICALLY MONITOR SENSOR-BASED SYSTEMS

摘要

Application specific tuning has been shown to be beneficial for a variety of platforms, sensor-based systems are no exception. However, accurately capturing external stimuli or modeling application specific stimuli remains a challenge. Tuning a sensor-based system to erroneous or incomplete application information can limit the optimization achieved or even negatively impact the resulting system performance. We propose to obtain accurate application stimuli by dynamically observing the sensor-based system in the intended deployment location, storing application characteristics for later analysis or optimization. A packet-based profiling methodology is introduced incurring overheads ranging from 0.07% to 27.8%, depending on the amount of sensor activity within the deployed system. Additionally, the profiler module and proposed packet-based profiling methodology have been implemented on the eBlocks platform to demonstrate the feasibility of the approach.