Dynamic power management in hard real-time systems

The approaches to energy minimization that are described in this thesis are based on the relatively new paradigm of dynamic power management (DPM) that has emerged in recent years. The most effective DPM techniques for energy reduction are based on run-time variation of processor voltage to reduce processor power, which is referred to as dynamic voltage scaling (DVS), and dynamic switching of I/O device states to minimize I/O power. In this thesis, both these approaches are shown to be extremely effective for real-time systems.; The thesis first addresses the problem of energy minimization of the processor by discussing the implementation of a low-energy task scheduler on a real-time testbed. The testbed consists of a real-time operating system running on a processor that can operate at a small set of discrete voltages. The evaluation of this online scheduling heuristic demonstrates the effectiveness of the approach in terms of energy savings and scheduler latency.; The thesis also demonstrates the application of network flow theory to the DVS problem. The DVS problem is modeled as a generalized network flow graph for processors that can operate at multiple speeds, and also as a minimum-cost network flow graph for multiple-speed processors.; The application of DPM techniques for I/O devices in hard real-time systems is also investigated in this thesis. Online and offline algorithms that perform run-time switching of device states to minimize energy consumption of I/O devices in real-time embedded systems are developed.; Finally, the application of network flow theory to precedence-constrained scheduling in distributed systems is studied. A generalized network flow model is developed for scheduling tasks with precedence constraints in a distributed system in which each processor is capable of operating at several speeds. (Abstract shortened by UMI.)...