SensorMash: Exploring System Fidelity Through Sensor Mashup

Context-aware services are driven by sensed data from both real and virtual worlds. Building effective pervasive systems involves administration and fine tuning of sensors toward optimal operation. Difficulties arise because sensors are prone to inaccuracies through miscalibration, malfunction and component limitations. Any incorrect values of sensed data need to be accounted for and dealt with appropriately otherwise the system as a whole may not behave in a useful manner. Trying to understand the limitations of sensors, service tolerances to inaccuracies, and the multiplicative effects of erroneous data on a given system can be an extremely complicated task. In this paper we present SensorMash, a tool for exploring sensor interactions as a mashup of inputs into a context-aware system. Built on top of a general model for sensor data, SensorMash allows developers to explore the effects of massaging tolerances to inaccuracy ratings and uncertainty. A small user trial is described with initial results driving research into an autonomic sensor management system.