Department of Computer Science & Engineering

Power consumption is a major concern in embedded systems design due to the portability and battery driven operation of such systems. The runtime optimization of embedded software pplications for system-level power / performance tradeoffs requires ability of the runtime system to probe system and application status and utilize procedures that make these tradeoffs effective. To ensure efficiency of decision making, it is important that such decisions are made with the least overhead to system power. One way to achieve this capability is through systematic definition, and update of meta data that can be probed by the runtime system and given as input to the dynamic power management algorithms. In this paper, we use the concept of application reflection, a technique in which a program represents its own structure and behavior through the use of meta-data. Its use enables the ability of the runtime system to look at the program representation and make power management related decisions. We present a profiling scheme to build a reflexive data structure in which a program represents its own execution behavior, and use this information at run time to guide operating system power management decisions. Our scheme is inspired on {\it Simpoint}, a tool for automatic program phase classification and simulation points selection. We use main memory bank shutdown as an example of how our technique can be used and we show that we can achieve energy/delay savings comparable to the best known hardware based technique. We believe that our approach can also be used for efficient energy management of other resources such as processor and system peripherals.

Pre-2018 CSE ID: CS2004-0797