Dynamic adaptation for fault tolerance and power management in real-time embedded systems

Safety-critical embedded systems often operate in harsh environmental conditions that necessitate fault-tolerant computing techniques. In addition, many safety-critical systems execute real-time applications that require strict adherence to task deadlines. These embedded systems are also energy-constrained since system lifetime is determined largely by the battery lifetime. We investigate dynamic adaptation techniques for real-time embedded systems, based on checkpointing and dynamic voltage scaling (DVS) for fault tolerance and power management, respectively. This thesis provides an integrated framework for achieving fault tolerance and energy savings in real-time embedded systems. Both hard and soft real-time systems are addressed in this work.; The thesis first proposes a unified approach for achieving fault tolerance and energy savings in hard real-time uniprocessor systems. Feasibility-of-scheduling tests are presented for checkpointing schemes for a constant processor speed as well as for variable processor speeds. DVS is then carried out on the basis of the feasibility analysis. Important practical issues in checkpointing and DVS are incorporated in the analysis. Simulation results based on real-life checkpointing data and processor models show that compared to fault-oblivious methods, the proposed approach significantly reduces power consumption and guarantees timely task completion in the presence of faults.; Next, this thesis investigates fault tolerance and power management in distributed real-time embedded systems. An application program is modeled by a directed acyclic graph where nodes correspond to jobs, and edges denote communication cost and precedence constraints. Coordinated checkpointing is used to achieve consistency. Feasibility analyses are carried out by examining each job's worst-case finish time in the presence of faults. Simulation results based on the CORDS hardware/software co-synthesis system highlight the efficiency of this approach for fault tolerance and power management.; This thesis also presents an adaptive checkpointing scheme in which the checkpointing interval for a task is dynamically adjusted during execution, and checkpoints are inserted based not only on the available slack, but also on the occurrences of faults. Adaptive checkpointing is then combined with DVS to achieve power reduction. Following this, the adaptive checkpointing scheme is extended for a set of multiple tasks in real-time systems. An off-line pre-processing step based on linear programming is used to determine the parameters that are provided as inputs to the on-line adaptive checkpointing procedure. A competitive ratio analysis is presented to evaluate the effectiveness of the proposed on-line scheme by comparing it against an optimal off-line scheme that has prior knowledge of fault occurrences.; Finally, the tradeoffs between real-time responsiveness and energy savings are investigated for soft real-time embedded systems. A new metric termed energy/real-time ratio (ERR) is proposed to evaluate the performance in terms of timeliness and energy savings.; In summary, this thesis provides an integrated approach for achieving fault tolerance and energy savings in real-time embedded systems. It trades off real-time responsiveness with fault tolerance and energy savings in a unified manner.