Abstract
In an IoT (Internet of Things) system, the autonomy of battery-operated edge devices is of paramount importance. When such devices operate intermittently, reducing power consumption during standby improves such a characteristic. The deep-sleep operation mode obtains such a result: it keeps on power only the hardware needed to wake up the unit after a timeout or an external trigger. For this reason, deep sleep exhibits the issue of losing the working memory, which prevents its use with applications depending on long-lasting or stateful computations. A way to circumvent such an issue consists of saving a snapshot of the working memory on a remote repository. However, such a solution is not always convenient since it exhibits an energy footprint due to checkpoint transmission. This article analyzes the applicability of such a solution. Firstly, by comparing its energy footprint against keeping the working memory on power. The analysis follows a formal, technology-agnostic methodology based on a mathematical model for energy consumption. It yields a discriminant inequality identifying the use cases where remote checkpointing is of interest. Once justified the approach, the article proceeds by defining an architecture and a secure protocol for data transport and storage. Finally, the description of a prototype implementation provides concrete insights.