Scalability and Robustness of Time-Series Databases for Cloud-Native Monitoring of Industrial Processes

Today's industrial control systems store large amounts of monitored sensor data in order to optimize industrial processes. In the last decades, architects have designed such systems mainly under the assumption that they operate in closed, plant-side IT infrastructures without horizontal scalability. Cloud technologies could be used in this context to save local IT costs and enable higher scalability, but their maturity for industrial applications with high requirements for responsiveness and robustness is not yet well understood. We propose a conceptual architecture as a basis to designing cloud-native monitoring systems. As a first step we benchmarked three open source time-series databases (OpenTSDB, KairosDB and Databus) on cloud infrastructures with up to 36 nodes with workloads from realistic industrial applications. We found that at least KairosDB fulfills our initial hypotheses concerning scalability and reliability.

I. Introduction

To supervise industrial processes, such as power production, oil refineries, or chemical processes, today's industrial control systems store large amounts of senso data into special databases. These databases then allow human operators to analyze trends in sensor data (e.g., temperature or flow pressure curves) and to optimize the industrial processes.