Packaging predictable assembly

Abstract

Significant economic and technical benefits accrue from the use of pre-existing and commercially available software components to develop new systems. However, challenges remain that, if not adequately addressed, will slow the adoption of software component technology. Chief among these are a lack of consumer trust in the quality of components, and a lack of trust in the quality of assemblies of components without extensive and expensive testing. This paper describes prediction-enabled component technology (PECT). A PECT results from integrating component technology with analysis models. An analysis model permits analysis and prediction of assembly-level properties prior to component composition, and, perhaps, prior to component acquisition. Analysis models also identify required component properties and their certifiable descriptions. Component technology supports and enforces the assumptions underlying analysis models; it also provides the medium for deploying PECT instances and PECT-compliant software components. This paper describes the structure of PECT. It discusses the means of establishing the predictive powers of a PECT so that consumers may obtain measurably bounded trust in both components and design-time predictions based on the use of these components. We demonstrate these ideas in a simple but illustrative model problem: predicting average end-to-end latency of a 'soft' real time application built from off-the-shelf software components.