UC San Diego

The goal of this dissertation is to introduce application adaptation techniques to maximize the quality of complex applications when certain hardware flaws exist in a platform. Two types of hardware flaws have been studied in this research: static hardware flaw, which is due to process variations, and dynamic hardware flaw, which is due to dynamic thermal management. The adaptation idea is based on the premise that complex and real-time applications provide a set of parameters that can be used to tune their complexity and quality. In this dissertation, we use this application layer opportunity to adapt applications to the effects of hardware drawbacks by tuning their parameters to achieve maximum quality. We then extend our solution and show how our proposed application layer solution can be combined with conventional hardware solutions to provide a hybrid adaptation method to tackle hardware drawbacks. The experimental results on practical platforms such as MacBook Air show that using our application adaptation can significantly improve the quality of real-time applications, such as video encoder and turbo decoder, while not requiring any modification to the hardware. In addition, the results show these adaptation methods can be used to reduce the cooling efforts such as processor fan speed