Experiences in implementing an energy-driven task scheduler in RT-Linux

Dynamic voltage scaling (DVS) is being increasingly used for power management in embedded systems. Energy is a scarce resource in embedded real-time systems and energy consumption must be carefully balanced against real-time responsiveness. We describe our experiences in implementing an energy driven task scheduler in RT-Linux. We attempt to minimize the energy consumed by a taskset while guaranteeing that all task deadlines are met. Our algorithm, which we call LEDF, follows a greedy approach and schedules as many tasks as possible at a low CPU speed in a power-aware manner We present simulation results on energy savings using LEDF, and we validate our approach using the RT-Linux testbed on the AMD Athlon 4 processor. Power measurements taken on the testbed closely match the power estimates obtained using simulation. Our results show that DVS results in significant energy savings for practical real-life task sets. We also show that when CPU speeds are restricted to only a few discrete values, this approach saves more energy than currently existing methods.