Components of IoT setups are subject to hardware failures, independently of software quality. These errors can be addressed by introducing hardware-level fault tolerance. This requires additional hardware components and the orchestration of tasks between them.

Erlang and Elixir, when running on BEAM, solve this problem for server-grade computers. However, IoT systems can also contain many microcontrollers that are too limited in resources to host BEAM.

AtomVM is a novel runtime for languages that traditionally run on BEAM. It brings many features of the runtime to embedded devices, including those necessary to monitor the failure of remote nodes and build fault-tolerant systems. As memory and computational power on microcontrollers are scarce, gauging how these features perform is essential.

In this study, we measure the cost of fault tolerance by comparing a single-node implementation of a particular setup with that of a fault-tolerant one. This test system handles the reception and processing of radio messages, a typical task for the embedded components of an IoT deployment.

Our measurements should help developers design applications that, while building on Erlang’s fault tolerance, fit the limited resources available on microcontrollers.