Energy-Aware Paired Sampling-Based Decision Model for Dynamic Mobile-to-Mobile Service Offloading

Power models are a critical element in current research regarding the effect of program-offloading decision making on the energy consumption of mobile devices. Several utilization-based power models have been proposed for measuring the energy consumption of locally running programs. However, the main challenge of utilization-based methods is that the models must be retrained for program units that use hardware components not addressed in the training phase. This paper proposes a paired sampling-based power model to address this critical challenge. The proposed power model estimates the energy consumption of an OSGi service asynchronously invoked in a multithreading environment on the basis of the overall remaining battery energy information at runtime without a connected power meter or energy profile for each specific hardware component of different devices. On the basis of the power model, an offloading decision model is proposed to dynamically determine whether a service invocation should be offloaded to a nearby mobile device over Bluetooth to conserve energy. The proposed approach was experimentally assessed regarding the correctness of decision making, energy gained by offloading service invocations, and weighted absolute percentage error of the estimated energy consumption compared with actual one.