Performance Modeling for Software Integration

Software construction with reusable components and legacy code has proven useful for fast and low-cost software development in many real world applications. However, since this methodology is often applied to domains with stringent performance and resource constraints such as embedded systems, its ignorance of non-functional constraints makes it very difficult and expensive to meet the constraints at a post-integration stage. In this paper, we propose a new methodology for (i) integrating a performance model with functional models during the design phase, and (ii) using the performance model parameters along with functional components throughout all SW development phases that usually rely on different tools. The performance model is constructed by examining the performance requirements and characteristics of functional components and their interaction patterns, and by associating them with functional components and patterns as performance parameters. The advantage of such modeling lies in that the performance model is easily integratable with functional components and can be reused just like reusable functional components. The proposed performance modeling, therefore, solves the cross-cutting issues of performanceaware functional design, and makes the consistent performance information available across the different phases of software lifecycle. The main technical issues for such performance modeling include how to partition the performance model along the function models and software architecture, how to determine and systematically measure the performance parameters of software components in each partition for reuse in constructing different performance models, and representing the performance parameters for use across different tools, and building a performance model for designtime analysis and runtime QoS management.