Thermal-aware periodic resources
Masud Ahmed · Human Biology · 2010
Over the years, thermal-aware designs have become a prominent research issue for real-time application development. A drastic increase in energy consumption of modern processors makes these devices more prone to overheating. This excess heat decreases the life span of devices running real-time applications. As a result, obtaining a reduction in peak temperature is considered the most desirable design aspect in developing such devices as it not only decreases the packaging cost, it also increases the lifetime of a device substantially. In this thesis, thermal-aware periodic resource model is proposed which is a proactive scheme for minimizing peak temperature in a system with a microprocessor having basic energy saving features. For this model, polynomial-time algorithms are proposed for determining lowest processing time (bandwidth) in periodic intervals with the minimum peak temperature for a specified sporadic task system scheduled by earliest-deadline first (EDF). Proposed algorithms only incur a small relative error in exchange for a significant improvement in running time over exact algorithms. We, furthermore, derive the thermal equations for calculating the asymptotic temperature bound for a given thermal-aware periodic resource. These equations, along with the algorithm presented, will give a system designer not only a guarantee of schedulability for a given workload, but also the freedom of choosing a tradeoff between the system utilization and the peak temperature.