Model-Driven Development of Self-organising Control Applications

Abstract

In this article we present a software development methodology and a supporting algorithm toolbox for Organic Computing applications that were developed in the context of the MODOC project. We focus on control applications for dynamic actuator/sensor networks that have no skilled human administrator and that are subject to frequent random hardware failures. To deal with these characteristics, applications in this setting should be self-organising and self-stabilising. However, realising self-organising and self-stabilising applications is usually a complex task requiring expert knowledge. Thus, we show that the development of such applications can be greatly simplified by shifting complexity into a model-driven tool chain. Developers implement their application in a high-level language based on a role abstraction focusing on application functionality. In order to hide the remaining complexity, we developed a tool chain which is responsible for transforming the high-level code to executable machine code that is automatically enriched by self-x properties such as self-organisation and self-stabilisation, which are not explicitly defined by the application programmer. Thereby, the application benefits from extended fault-tolerance and availability. A major part of the tool chain is a toolbox containing self-x algorithms commonly required for Organic Computing applications. The heart of the toolbox is an algorithm stack realised by our publish/subscribe middleware Rebeca that is able to efficiently realise self-organisation even in case of large networks.