New Scheduling Strategies Boost Batteryless IoT Reliability for Unknown Workloads

Batteryless Internet of Things (IoT) devices, which rely on energy harvesting, face significant challenges in maintaining reliable operation due to highly volatile energy storage and unpredictable workloads. Traditional energy-aware schedulers often fall short because they depend on static execution thresholds or pre-measured, hardware-specific task profiles, which are impractical for complex, unknown edge applications. To address this, a new study introduces two dynamic scheduling strategies that operate without prior energy information, treating applications as a "black box." These include a model-free Reinforcement Learning (RL) agent and an on-the-fly Approximated Prediction (AP) method. Evaluations using a custom simulation framework, driven by real-world solar data, revealed distinct trade-offs: the AP approach offers lightweight, near-optimal task throughput, the RL agent provides a balance between system survival and task execution, and an adaptive task rate (AsTAR) method excels during long energy gaps. The research also indicates that while these advanced strategies are crucial for severely constrained systems, devices with larger energy buffers might still effectively use simpler static policies.