Thermal-Constrained Energy Efficient Real-Time Scheduling on Multi-Core Platforms

As the power density of Integrated Circuit chips continues to grow, energy minimization and temperature reduction are two of the most critical design issues in today’s computer system design. Although energy minimization is closely related to temperature reduction, the most energy-efficient method may not be the most effective one to meet the temperature constraints and vice versa. In this paper, we study the problem of how to partition periodic hard real-time tasks on a multi-core platform to maximize the overall energy efficiency under a peak temperature constraint. Differing from the traditional load-balancing approach, i.e., evenly distributing the workload across the chip, we propose a thermal-balancing strategy, i.e. minimizing the thermal gradient across the active cores, to improve the overall system energy efficiency, especially when the temperature constraints are tight. We first identify the lower bound for energy consumption with this approach, and then transform the task partitioning problem to a variable-sized bin packing problem. We further propose an enhanced algorithm to optimize the task partitioning results. Our simulation results show that the proposed thermal-balancing approach can significantly improve the energy efficiency and task partitioning feasibility for real-time systems with high system utilization and tight temperature constraints.