Self-adaptive distributed MQTT middleware for edge computing applications

Abstract

In recent years, Edge Computing has emerged as a new computing paradigm that leverages computing resources close to clients, paving the way for latency-sensitive applications like industrial IoT or urban augmented reality. Many of these applications use publish-subscribe patterns for communications that use message brokers for data dissemination. Currently, most message brokers focus on deployments in the cloud. However, the connections between clients and cloud brokers introduce latencies that some applications might not tolerate. Edge computing resources offer a possible location to deploy brokers closer to clients, reducing latency. We designed a distributed broker system that makes use of edge resources to improve Quality of Service. Our approach consists of proximity detection by using a network coordinate system, along with self-adaptive broker orchestration. Using network coordinates allows us to estimate distances instead of having to measure them, reducing monitoring overhead and network strain. The broker orchestration starts and stops broker instances on edge resources on demand, using local pressure on edge nodes by close clients to decide whether to start a broker or not. We evaluate our system in a simulated environment with scenarios and topologies that are representative for IoT applications. Results show that we can use a network coordinate system to accurately find the closest broker for a client, and at the same time reduce the number distance measurements by at least one order of magnitude. We also show that the broker scaling mechanism dynamically deploys brokers on the edge resources close to the clients, while taking the tradeoff between end-to-end latency and resource consumption into account.