Benchmarking WebAssembly for Embedded Systems

WebAssembly is a modern, low-level virtual machine with designed for improved application performance in web browsers. Recently, WebAssembly gained interest for its use outside the web, for example as a replacement for serverless container runtimes. A number of non-web WebAssembly implementations are actively supported, some of which target microcontrollers, IoT devices, and embedded systems. Such hardware platforms have strict resource constraints which may render the usage of WebAssembly impossible or too costly, for example, due to its performance overhead and memory requirements. However, it is currently unclear what performance to expect of WebAssembly on low-resource microcontrollers compared with machine code and alternative application virtual machines. To answer this question, we evaluated the processing overhead and memory characteristics of WebAssembly application virtual machines on microcontrollers, and compared it to native execution, and the established application virtual machines: MicroPython and Lua. Furthermore, we analyzed the feature-set and architecture of the WebAssembly implementations in more detail, and measured the performance impact different runtime features have. We found that WebAssembly, despite its high extensibility and versatility in supported source languages, application paradigms, and target hardware, delivers very competitive performance. We conclude that WebAssembly can find wider industry usage for embedded systems and could replace other more costly or less flexible virtualization techniques, such as Java.