Low Power Software For Multiprocessor Systems

An important issue in the development of mobile embedded systems is the optimization of energy consumption. The increasing complexity of these systems leads to an increased power consumption. Unfortunately, the battery capacity grows at a smaller rate than the power requirements from the applications, leaving a gap that must be filled by various system-level optimizations. Dynamic voltage scaling and adaptive body-biasing are well-known techniques to reduce dynamic and leakage energy effectively. These voltage settings can be calculated either offline, before the actual execution of tasks or online, during runtime. The advantage of online optimizations is the utilization of dynamic slack that results from variations in the execution time. However, these calculations are expensive and time-consuming. This book introduces a quasi-static algorithm for combined supply voltage scaling and body biasing. Voltage settings are computed offline and readjusted during runtime with very low online time complexity. This work considers time-constrained and discrete multiprocessor systems as well as energy and time overheads for the recalculation and readjustement of the voltages.