Dynamic IoT Applications and Isomorphic IoT Systems Using WebAssembly

With the proliferation of Internet of Things (IoT) devices, developers need to respond to diverse and rapidly changing user requirements quickly. Thus, a method for rapidly and frequently updating IoT devices is required. Moreover, owing to the development of cloud/edge computing, an approach to enable the efficient development and maintenance of multi-layer IoT systems is necessary. In this study, we propose a method for building dynamically updatable IoT devices using WebAssembly (Wasm), where the IoT device consists of a combination of the primary programming language implementing the application and the Wasm runtime. Furthermore, we propose isomorphic IoT systems that use the same Wasm binary that is built from a common code base in each layer of the systems. We showed a specific use case of compiling machine learning models for image recognition and classification into Wasm binaries and built an isomorphic architecture where the inference process is executed using the same Wasm binary in each layer of the IoT system. We performed a quantitative evaluation to confirm the effectiveness of the proposed method. Although the proposed method introduces an overhead that is caused by calling Wasm functions from the application, the impact thereof is limited. We also measured the performance of the proposed method using the widely used image recognition and classification models ResNet-50 and MobileNetV2. We confirmed that the proposed method is practical in the current situation and offers promise for the future.