System-Level Design Methods for Low-Energy Architectures Containing Variable Voltage Processors

Energy aware scheduling for distributed real-time systems

“…They improved the static optimization by using critical path analysis and task execution order refinement to get the maximal static slow down factor for each task [16]. For a fixed task set and predictable execution times, static power management (SPM) can be accomplished by deciding beforehand the best voltage/speed for each processor [8]. When there are dependence constraints between tasks, for a given task assignment, Gruian et al proposed a priority based energy sensitive list scheduling heuristic to determine the amount of time allocated to each task, considering energy consumption and critical path timing requirement in the priority function [9].…”

“…Following the same idea, a simple static power management (S-SPM) scheme for multiprocessor systems was proposed by distributing global static slack over the length of a schedule [8]. Applying S-SPM to the task graph in Figure 1a, we see that every task will run at 2 3 f max and the static schedule is shown in Figure 2b.…”

Leakage Temperature Dependency Aware Real-Time Scheduling for Power and Thermal Optimization

“…The energy-efficient scheduling in multi-core systems are often NP-hard, as a result several heuristics and approximation techniques are studied to minimize dynamic energy consumption on multi-core platforms [143,33,50,148]. In [50] Gruian used simulated annealing algorithm for task to core allocation and developed two-stage energy minimization method when scheduling tasks with inter-dependencies.…”

“…In [50] Gruian used simulated annealing algorithm for task to core allocation and developed two-stage energy minimization method when scheduling tasks with inter-dependencies. Zhang et al [143] proposed a technique that finds the optimal task allocation for energy minimization by exhaustively checking all possible permutations of task allocation running on the minimum speed that could guarantee real-time constraints for each task.…”

“…The running mode for each scheduling interval was randomly chosen from [0, 0.8, 0.9, 1.0]V (see Table 6.2(a)). The total length of the schedule interval was evenly distributed within [100,200], and the length of each scheduling interval was evenly distributed within [50,80]. For each test case, our proposed method as well as the traditional numerical method with sampling interval varies from 0.01 second to 1 second were used to calculate the energy consumption.…”