Component deployment optimisation with bayesian learning

Abstract

Implementing embedded software systems involves many important design decisions, such as finding (near) optimal component deployment architectures, that have a strong influence on the quality of the final system perceived by its users. These decisions are difficult not only because of the complexity of current systems, but also due to the large number of possible design options. An automation of the design space exploration will help to make better decisions and to reduce the time of this process. In this paper, a new method called Bayesian Heuristic for Component Deployment Optimisation (BHCDO) is proposed. BHCDO constructs solutions based on a Bayesian learning mechanism which guides the search for assignments that result in new deployment architectures with an improved quality. This is achieved by calculating the posterior probability that a particular component/host assignment is a good decision, resulting in a high quality deployment architecture, given some observed evidence during the search. Experiments on a series of randomly generated problems show that BHCDO efficiently automates the search for component deployment design alternatives and outperforms state of the art optimisation methods.