Quality-Driven Dynamic Scheduling for Real-Time Adaptive Applications on Multiprocessor Systems

While quality-adaptable applications are gaining increased popularity on embedded systems (especially multimedia applications), efficient scheduling techniques are necessary to explore this feature to achieve the optimal quality output. In addition to conventional real-time requirements, emerging challenges such as leakage power and multiprocessors further complicate the formulation and solution of adaptive application scheduling problems. In this paper, we propose a dynamic adaptive application scheduling scheme that efficiently distributes the runtime slack to achieve maximized execution quality under timing and dynamic/leakage energy constraints. Our proposed methods are threefold: First, for each task in the slack receiver group, a heuristic guided-search algorithm is proposed to select the optimal processor frequency to maximize the application execution quality. Second, we present an efficient slack receiver selection methodology aiming at identifying optimal slack receivers for quality maximization. Third, our framework is further extended to consider constraints brought by interprocessor communications, where we study the effects of slack inaccuracies introduced by transmission variations, and propose a local scaling approach to compensate the induced quality loss. Experimental results on synthesized tasks and a JPEG2000 codec show that the guided-search algorithm, aided by slack receiver selection, effectively outperforms contemporary approaches with at most 88 percent more quality improvement, whereas the local scaling contributes as large as 16.9 percent on top of the guided-search results.