MQTT Tuning at the Edge: Resource-Aware Optimization of MQTT Communication for Edge IoT

Modern edge IoT platforms are relying more and more on the Message Queuing Telemetry Transport (MQTT) protocol for local low-latency data exchange between sensors, gateways, and control applications. The performance of a broker running on such a resource-constrained platform is greatly impacted by decisions made on Quality of Service (QoS) level, payloads, rates, number of clients, and transport encryption. These decisions are often counter intuitive and have to balance the CPU, memory, and network usage, introducing operational overhead. This paper proposes a novel methodology for resource-aware tuning of MQTT communications that can be applied to any MQTT-enabled edge platform. The methodology was tested on an air-gapped edge IoT platform, implemented on a Raspberry Pi 5 node with Mosquitto broker. Key parameters in the MQTT communication were isolated by performing a series of emqtt-bench workloads, measuring the performance of a broker using Prometheus/Grafana monitoring, and identifying trends between the QoS levels, payloads, number of clients, rates, and TLS-level. The proposed approach does not attempt to find a single best point but rather defines a region of interest in the performance-resource space, wherein a practitioner can look for parameters that would best fit their use case. Trends are presented as hypotheses to be validated on the testbed and the operational overhead of the observability is also modeled as a part of the methodology.