Design and implementation of an advanced MQTT broker for distributed pub/sub scenarios

MQTT is one of the most popular communication protocols for Internet of Things applications. Based on a publish/subscribe pattern, it relies on a single broker to exchange messages among clients according to topics of interest. However, such a centralized approach does not scale well and is prone to single point of failure risks, calling for solutions where multiple brokers cooperate together in a distributed fashion. In this paper, we present a complete solution for a distributed MQTT broker systems. We target several functional primitives which are key in such a scenario: broker discovery and failure recovery, overlay tree network creation and message routing. Moreover, we also focus on the case where multiple topics are present in the system. In such a scenario, a single tree-based overlay network connecting the different brokers may not be the most efficient solution. To cope with this issue, we propose a topic-based routing scheme for MQTT distributed brokers. The proposed solution creates multiple overlay networks in the distributed system, each one linking together only the brokers whose connected clients have interest in the same topics. We implement the complete system as an extension of the popular HiveMQ MQTT broker and perform several experiments to test its performance in scenarios characterized by a different publishers/subscribers configurations as well as number of topics existing in the system.