System level energy management in networked real-time embedded systems

Real-time embedded systems play a prominent role in a variety of applications ranging from medical sensors in human body to signaling sensors in war fields. The consumer domain of the embedded devices is large and ever increasing. A natural result of this trend coupled with those in sensor technologies and wireless communications have led to the rise of a new class of systems, called the networked real-time embedded systems. While the networked embedded systems enable newer and functionally richer applications, they pose major research challenges due to the complex requirements of satisfying temporal and reliability constraints in a resource limited distributed computing environment. This dissertation develops a comprehensive algorithmic framework for system-level energy management in networked real-time embedded systems, with the goal of optimizing the energy consumption while satisfying application requirements (deadlines and precedence relations) and channel reliability constraints. The energy-management problem is decomposed into three levels and research contributions are made for each level: computing subsystem level, communication subsystem level, and the system level. For the energy management at computing subsystem level, cross-layer energy-aware task scheduling algorithms are presented which employ the dynamic voltage scaling (DVS) power management technique to minimize the processor energy consumption while meeting all the task deadlines. Simulation studies show that the presented cross-layer algorithms yield enhanced processor energy savings compared to the existing algorithms for a variety of workload conditions. The communication energy management is addressed considering two different power management techniques namely, dynamic modulation scaling (DMS) and power adaptation. In each case, the energy-aware message scheduling problem is formulated tackling which both analytical and algorithmic solutions are presented. Performance results show that the proposed polynomial-time heuristic scheduling algorithms offer comparable energy savings to that of the analytically derived optimal solutions. Thirdly, system-level energy management is addressed for a model where the individual nodes in the network support both DVS and DMS techniques. This work is the first of its kind that combines compute-level and communication-level energy management in an integrated manner. Tradeoffs between compute and communication energy consumption is established and cut-off region that favors one over the other is derived based on task/message and system characteristics - current processor frequency, current radio modulation level, task deadline, channel bandwidth, source-destination distance. Further, specific system-level energy-aware scheduling problems are formulated for single-hop and multi-hop networked embedded systems with deadline constraints. For the problems, based on the above system-level tradeoff analysis, novel algorithms for combined optimization of computation and communication energy are presented. Our performance results show that the proposed system-level energy-aware algorithms perform significantly better than the corresponding component-level algorithms. In conclusion, this dissertation advances the state-of-the-art research in energy management in networked real-time embedded systems through an integrated framework and associated cross-layer algorithms and analyses. The system-level energy management formulation provides several avenues for further research, which include instantiating different power management techniques for computing and communication energy optimizations and studying their tradeoffs