Microservices in IoT clouds

The current Internet of Things technology landscape is admittedly plagued with fragmentation. Fragmentation in IoT seems to be abundant ranging from the device hardware, operating system and software to device-to-device and device-to-cloud-based infrastructure protocols, to the actual cloud-based infrastructure and tools used for developing and operating software that runs on these two opposite ends of the system. As such it is challenging for a developer to decide the development, deployment and operational technologies for a complete end-to-end IoT solution. In this paper we focus on the performance evaluation and technology selection of the cloud end of the system. Given that a modern IoT cloud infrastructure is based on a microservices platform, we tackle the challenge of the selection of a Microservice Application Server (MAS) among several available in the developer community for the task of data collection from IoT devices. This paper provides two key contributions. First of all we present an empirical evaluation between different JVM-based MASs and corresponding client/frameworks taking into consideration throughput (req/s), memory footprint and binary footprint. Secondly we provide an open-source testbed that can be used for reproducing the current evaluation and for extending the evaluation towards additional MASs that have been implemented in different runtimes/programming languages.