The on-board computer system of the GeoSurv II Unmanned Aerial Vehicle (UAV) is a small-scale loosely coupled distributed real-time and embedded (DRE) system. Ideally, a loosely coupled DRE system is combined with applications, a publish/subscribe middleware, a real-time operating system and time deterministic network hardware. Even though many performance modeling and analysis approaches are available, there is still a need of a simple model to help application developers predict the performance of the GeoSurv II UAV computer system.
This dissertation addresses the need for the development of an analytic modeling approach named Path Network (PN) modeling, which can be used to predict whether a loosely coupled DRE environment is appropriate for the GeoSurv II UAV applications. The visual notation of the PN model is designed, and the temporal parameters of the PN model are characterized. The environment overheads are broken down into several pieces for performance prediction. The algorithm for calculating the performance is described in detail. The prediction results are compared with the measurement results of a UAV experimental system to validate PN modeling. Even though the test-bed is not an ideal loosely coupled DRE environment, the results of the case study still indicate that PN modeling provides a simple and easy way for application developers to predict the performance of the on-board, loosely coupled DRE system of the GeoSurv II UAV.
~
Path Network (PN) modeling is an approach to performance modeling and analysis in which an application is represented as a network of message-paths that are evaluated analytically. A PN model is the repository of all relevant structural elements and their relationships, as well as their timing properties such as the processing time and environment overheads. An outline of PN modeling is introduced in Section 5.1. How to use the visual notation of PN modeling to build a PN model is described in Section 5.2. How to measure the environment overheads is provided in Section 5.3. Section 5.4 provides the prediction algorithms for application developers to calculate the performance results. Finally, the methodology of PN modeling is summarized in the last section.