Checkpointing in Transiently Powered Sensor-Based Internet of Things (S-IoT) Networks

Abstract

The era of smart cities is upon us, requiring the deployment of large-scale networks where autonomy is essential. Crucial to this autonomy is a continuous energy supply. However, a major shortcoming in Sensor-Based Internet of Things (S-IoT) networks is the finite energy supply available for computation and communication. To address this issue, energy harvesting has been proposed to enable continuous energy supply to IoT devices. However, fluctuations in energy supply due to energy harvesting often lead to node crashes due to energy exhaustion. Then, applications lose their state and checkpointing in non-volatile memory (NVM) has been proposed to persist state across crashes, albeit at the expense of significant overheads. While checkpointing is popular in traditional distributed systems, the state-of-the-art for transiently-powered sensor-based IoT (TP-IoT) has focused on checkpointing in uniprocessor setting (i.e, on a single device). Therefore, evaluating checkpointing strategies in TP-IoT networks is essential. There are three main factors to consider when checkpointing is done: (i) when to checkpoint, (ii) what to checkpoint and (iii) when to restore. Our contributions are: (i) we run testbed experiments to understand when is checkpointing beneficial (or not) in a computation, (ii) we develop a framework that guides in the selection of variables for checkpointing and (iii) we reduce the checkpointing to a precedence-constrained scheduling problem, called the CheckIn problem, and propose an adaptive algorithm that outperforms existing checkpointing strategies.