Abstract
Edge computing attempts to deliver low-latency services by offloading data storage and processing from remote data centers to distributed edge servers near end users, whereas network protocols, designed for centralized management, do not internally scale to distributed edge scenarios. In this paper, we establish the message dissemination support of MQTT, a de facto protocol for Internet of Things, for fully distributed edge networks. We summarize and formulate existing mechanisms, namely publication flooding and subscription flooding, and propose a topic-centric solution called selective subscription forwarding, which forwards subscriptions only when necessary by leveraging the topic containment of MQTT messages and therefore reduces inter-broker traffics. Evaluation results demonstrate that compared with existing solutions, more than 40% subscription traffic can be reduced with the proposed mechanism.