MQTT Broker Optimization: Comparative Analysis of Round Robin and Least Response Time

Abstract

Optimizing MQTT broker performance is crucial for maintaining efficient message routing in IoT systems, especially under varying workloads and QoS levels. This study compares the Round Robin (RR) and Least Response Time (LRT) algorithms to evaluate their performance across QoS levels 0, 1, and 2 and client loads ranging from 500 to 2,500 clients. Using Apache JMeter, key metrics such as CPU usage, throughput, delay, jitter, and response time were assessed. LRT was found to excel in enhancing response time and reducing delay, particularly under high client loads and in applications requiring minimal latency. However, this comes at the cost of higher CPU usage under heavy loads. In contrast, RR demonstrated optimal performance in maintaining balanced CPU utilization and predictable performance, though with slightly higher response times. Both algorithms demonstrated linear scalability in throughput, confirming their ability to handle increasing workloads without bottlenecks. These findings offer practical guidance for IoT developers: in latency-sensitive environments such as industrial automation, LRT is preferable due to its low-latency benefits, while RR is better suited for resource-constrained IoT systems like environmental monitoring, where stability and even load distribution are prioritized. The trade-offs identified provide valuable insights for selecting appropriate algorithms based on specific application requirements.