A Control-Theoretical Approach to Adapt Message Brokers

Abstract

Messaging brokers allow the development of distributed applications whose components are weakly coupled and communicate asynchronously, e.g., publish/subscribe systems. Brokers usually have a set of configuration parameters (e.g., queue size, persistence mode) that can be configured to adjust the broker’s behaviour to an expected workload. However, developers only configure these parameters at deployment time and they are kept unchanged at runtime. This paper presents a step by step to implement a control-theoretical solution that dynamically changes messaging brokers’ behaviour. Central to the steps is the definition of controllers, along with their tune methods. Then, different controllers were implemented and tuned using distinct mechanisms following the proposed steps. The controllers have been integrated into a widely adopted open-source messaging service named RabbitMQ. The unique contribution of this paper is to show the steps needed for applying control theory to message brokers, along with a comparative analysis of different controllers and tuning methods. Developers can use similar steps in engineering control solutions for messaging systems.