Exploring performance trade-offs of a JPEG decoder using the deepcompass framework

Designing embedded systems for multiprocessor platforms requires early prediction and balancing of multiple system quality attributes. We present a design space exploration framework for component-based software systems that allows an architect to get insight into a space of possible design alternatives with further evaluation and comparison of these alternatives. The framework provides (a) tool-guided design of multiple alternatives of software and hardware architectures, (b) early design-time predictions of performance properties and identification of bottlenecks for each architectural alternative, and (c) evaluation of each alternative with respect to multi-objective trade-offs. The performance prediction technique employs modeling of individual components and composition of the models into a system model representing the system behaviour and resource usage. We illustrate the framework by a case study of a JPEG decoder application. For this system, we consider architectural alternatives, show their specification, and explore their trade-offs with respect to task latencies, resource utilization and system cost.