Mid Sweden University

In the last years, the industrial automation has experienced a deep transformation known as Industry4.0, and it is driven by Internet of Things (IoT) paradigm. The IoT-based automation is based on well-defined data models, which make easy the interaction among devices. Generally, the data generated by IoT sensors are elaborated to obtain value added services (such as predictive maintenance), exploiting cloud services and remote servers. An accurate timestamp of the data generated by sensors is required to maintain an adequate level of such services: an 'easy' task in the case of a new deployment, but a nightmare when existing plants or machinery are retrofitted. In this case, the data are timestamped at cloud level, using the remote time. In such situations, a knowledge of the sense of time of cloud services is fundamental to guarantee the quality of data elaboration. The target of the research is an experimental characterization and a comparison of time awareness of different commercial cloud service providers (i.e. Amazon AWS, Google Cloud and Microsoft Azure). The characterization highlights as, generally, the performance provided by different platform is comparable each other. The time offset of NTP (Network Time Protocol) clients running on different Virtual Machines (VMs) has an uncertainty ranging from 0.05 ms up to 0.6 ms depending by the client configuration. Such results demonstrate that extreme care must be taken when using the time of remote VMs. © 2019 IEEE.