Performance and reliability modeling for a parallel image fusion system

This paper presents a novel parallel algorithm for multispectral image fusion called the s-PCT. The algorithm is designed based on a manager-worker threading concept and goes further to use redundancy to achieve fault tolerance. The paper emphasizes asymptotic analysis using analytical models and empirical studies that allow performance to be assessed for a wide variety of typical variations in use. For example, changes to the number of spectra, image resolution, number of processors, degree of redundancy, and reliability. Software reliability concept is studied and applied to the algorithm analysis. The performance and reliability relationship is also defined. The motivation in deriving the performance-reliability model is to assess the impact of changes in technology and problem size associated with different applications, allowing performance-reliability tradeoffs to be assessed. In other words, given a problem size and a reliability constraint, the model can be used to estimate the achievable performance.