Efficient Storage Backends for IoT Data : Consistent global state for input and coordination data in the Internet of Things

Abstract: Storing information is at the core of almost every Internet-of-Things (IoT) middleware, which poses challenges relating to scalability, fault-tolerance and correctness. This is even more relevant for a subset of IoT platform architectures that use a database abstraction to connect IoT components and treat data that represents physical state, control output, or inferred information in a uniform manner. At the same time, new database products aim to address these challenges through means of distribution while retaining semantics similar to existing relational databases. This study analysed the requirements of a stateful IoT middleware on its database backend in order to identify viable candidates. A middleware product was adapted to store its data on two backend implementations based on the databases CockroachDB and Infinispan, which make use of different concepts and algorithms. To leverage their transactional capabilities, the frontend interface of the middleware was supplemented with transactional access methods. Finally, the performance of the implementations was measured and compared with a set of application-oriented benchmarks. The results show a considerable trade-off between consistency, transactional safety, performance, and fault-tolerance and suggest that IoT middleware should make a distinction between the processing of input data aggregation and decision-making coordination tasks.