NO FULL TEXT AVAILABLE. This thesis contains 3rd party copyright material. ----- Being able to predict the overall performance of a Computer Based System (CBS), prior to building it is one of the greatest challenges facing system engineers today. Unfortunately this is currently, arguably, something of a black art, relying on previous experience, local knowledge, and ad-hoc techniques. While good models for individual parts of many systems exist, an overall model has not yet been defined. The emerging discipline of system architecture, by defining a high level abstraction of the overall structure of the system, may provide a basis for this reasoning. In working at this high level of abstraction, a system architect is able to make fundamental design decisions, early in a project, that will affect system properties such as performance, evolvability and openness. Most importantly, it has been acknowledged that as performance is a function of the components and connections of a system, it is at the architectural level that the performance of a system can be analysed and predicted.
Currently, architecture-based performance analysis work is focussed on simulating and analysing Layered Queueing Network Models (LQNM) of system architectures. It is shown that LQNMs are not suitable for architecture-based performance analysis of Open Computer Based Systems (OCBS) primarily because of their inability to deal with shared resources. OCBSs are CBSs that through the use of public, consensus-based specifications or standards, are focussed on achieving the interoperability between, and the portability of, components. It is shown that as the specification and standardisation ofboth the connections and the components of the system are of critical importance to realising the advantages of OCBSs, a layered architectural design is undertaken. This layered architectural design makes OCBSs extremely suitable to an architecture-based approach to performance analysis.
An architectural model suitable for the performance analysis of OCBSs is developed, based on Activity Cycle Diagrams (ACD), that improves on the LQNM approach. The ACD based OCBS architectural model is proven to be well founded by transforming it into the existing well founded techniques of LQNMs and Timed Coloured Petri Nets (TCPN) and analysing the resultant transformations. Furthermore, the reasoning possible with the model is derived with formal analytical predictions for service, waiting and response times, bandwidths, queueing and device utilisations developed.
Finally, an experimentation and simulation-based process for predicting and assessing the performance of OCBSs is defined. The process populates, implements and analyses the OCBS Architectural Model and it is proven to accurately predict the performance of a real-world case
study, namely the COLLINS class submarine Open System Extension (COSE) Concept Demonstrator (CD) system.
Following a literature review of the fields of Computer Based Systems, architecture, performance analysis and Open Systems, the current state of performance analysis of OCBSs is established and it is shown that an architectural model that allows the performance analysis of them is required. This OCBS Architectural Model is then developed based on a simplified OCBS reference model and the techniques of LQNMs and ACDs and formally represented and described using the Architectural Description Language (ADL) ACME. The model is then proven to facilitate performance analysis by proving it is well founded and theoretical predictions for service, waiting and response times, bandwidths, queueing and device utilisations are derived. Finally, an architecture-based performance analysis process for OCBSs is developed and shown to accurately predict the performance of a real-world case study.