Evaluation of Autoscaling Metrics for (stateful) IoT Gateways

Abstract

Wide adoption of Internet of Things solutions in the industry requires IoT platforms to support long-running, stateful protocols with numerous connections to ingest the measurements, and at the same time to provide the highest throughput possible to achieve service level objectives (SLOs). The challenge of providing sufficient throughput can be solved via automated horizontal scalability for IoT platforms' gateways. Firstly, to address this challenge, a study of the state-of-the-art solutions and approaches adopted in the industry to ensure predictable throughput at IoT gateways was conducted. Secondly, the following three gateways autoscaling metrics were studied on the in-house developed cloud-native IoT Platform: CPU utilization, the number of concurrent active connections, and the absolute throughput per gateway. The results of the experimental evaluation suggest the combination of rate limiting and active connection monitoring as a prerequisite for high throughput at IoT gateways.