Thermal-Aware Scheduling of Critical Applications Using Job Migration and Power-Gating on Multi-core Chips

Multi-core System-on-Chip (SoC) has become a popular execution platform for many embedded real-time systems. As CMOS transistors continue to shrink down to the nanoscale regime, it becomes more susceptible to various reliability threats mainly due to thermal hotspots. To improve the reliability of embedded real-time systems, dynamic thermal management (DTM) is required for mission/safety- critical applications running on a multi-core chip to avoid possible thermal hazards while meeting the applications´ timing constraints. In this paper, we propose an efficient runtime thermal- aware scheduler (TAS) using job-migration and power-gating techniques to avoid the thermal hotspots on a multi-core chip. Before runtime, the TAS distributes the periodic real-time tasks to cores using the tasks´ execution profiles to balance the utilization of functional units on the chip. At runtime, the TAS periodically monitors the core temperatures and triggers one of the following pre-defined thermal management schemes depending on the level of the measured core temperature: (i) migrating jobs running on hot cores to other cooler cores to reduce the workloads on the hot cores, or (ii) turning off hot cores for a certain period of time to cool them down. All tasks´ timing constraints are guaranteed during the job migrations and powering cores on/off. Our in-depth evaluation has shown that the proposed TAS can effectively minimize the thermal hotspots on a multi-core chip without violating any application timing constraint.