Adaptive Battery Charge Scheduling with Bursty Workloads

Abstract-Battery-powered wireless sensor devices need to be charged to provide the desired functionality after deployment. Task or even device failures can occur if the voltage of the battery is low. It is very important to schedule the recharge of batteries in time. Existing battery scheduling algorithms usually charge a battery when its voltage drops below a fixed level. Such algorithms work well when the workloads are predictable. However, workloads of wireless sensors can be highly bursty, i.e., extensive sensing and communication tasks usually occur in a very short time period. If such a bursty workload occurs when the battery voltage is low, the battery energy can be depleted very quickly, resulting in system task failures before the device can be recharged. To deal with unpredictable bursty workloads, we investigate battery characteristics with different workloads via experiments. Based on the empirical results, we build an adaptive linear model and propose a feedback control based battery charge scheduling algorithm. T his algorithm dynamically adjusts the battery charge threshold for recharge scheduling, adapting to bursty workloads. We have tested our algorithms in extensive simulations with traces obtained from real experiments. Evaluation results show that our algorithms can adapt to bursty workloads. Compared to existing algorithms, our algorithm achieves a 68.26 % lower task failure ratio with a 3.45 % sacrifice on system lifetime under bursty workloads. Index Terms-battery, burstiness, scheduling, control, energy efficienc