Deployment Optimisation Insights for Evolving Software Systems within a Constrained Computing Environment

Abstract

For the successful integration of System of Systems (SoS) type designs, there is a critical need to understand the non-functional aspects of the design and how these design aspects are affected by new or evolved operational conditions and subsequent deployment scenarios. An undersea sensor network system is one example of an SoS design where non-functional design aspects, such as power consumption and temporal performance, are critical to the overall system performance and capability, and are largely influenced by high-level operational needs. There is a crucial need to develop methodologies for modelling and understanding non-functional properties and corresponding deployment scenarios early in the development and integration cycle. This includes the initial development and subsequent evolution upgrade cycles. This thesis presents literature that shows there are a small number of pockets of research looking into measurement-based techniques for performance prediction of SoS designs early in the development cycle. Furthermore, the literature shows that System Execution Modelling (SEM) techniques lead the way in providing low-level insight into non-functional requirement behaviours and performances. However, the techniques do not have sufficient capability for managing the complexities and scale of the deployment of the software that make up SoS designs. The literature also shows that there are optimisation techniques for problem solving for many computing paradigms, including software deployment. However, these approaches are not adequate for software component deployment. Furthermore, the literature shows system evolution performance prediction techniques are either postmortem-based or largely abstract in nature, and not to the level of fidelity required for SoS performance insights. This thesis details new research to address the capability gap for modelling deployment of large-scale evolving SoS. It will introduce a new software deployment optimisation technique and modelling language that allows for evolution characteristic definition and construction of evolved designs. It will present a software deployment optimisation technique, algorithms and frameworks, and a capability that enables software deployment to be shaped by high-level requirements. We demonstrate the new optimisation capability with an undersea sensor system case study, where we explore design and deployment options for achieving highlevel defined performances. The testing and analysis presented in thesis shows the new optimisation approach operates and predicts correctly for each element of its optimisation algorithm. While some performance shortfalls exist, largely due to the model fidelity improvement requirements, our extensive verification and validation indicates the correct overall performance and ability to identify software component deployment options in response to high-level requirements. To complement the new optimisation capability, this thesis also introduces a new DSML to allow for exploration of software deployment optimisation for evolving systems. Known as CEML, this new language allows system designers to gain insight into how best to utilise its available computing resources when the software system evolves. The application of this new language was demonstrated by working through an undersea sensor system evolution case study. This thesis introduces four contributions to the research community associated with modelling and predicting system design performance, including system evolution.