Energy- and Temperature-aware Scheduling: From Theory to an Implementation on Intel Processor.

Temperature, energy, and performance are the key considerations of real-time multicore systems. Thermal hotspots and high temperatures not only degrade reliability and per-formance but also increase cooling costs and leakage current. To address these issues, we propose an energy, performance, and temperature-aware semi-partitioned scheduling technique for sporadic real-time tasks. The proposed approach incorporates DVFS, Dynamic Power Management, and chip floor-plan in clas-sical partitioned scheduling technique. To further reduce temper-ature, energy, hotspots, and gradients, the proposed partitioned scheduling also considers individual task behaviors in the form of hot and cold tasks and applies reactive thermal management techniques such as task migration (hence called semi-partitioned scheduling) for system safety. To our knowledge, this is the first paper that considers energy, performance, reliability, and temperature together in real-time multicore scheduling. Finally, we present the results through experiments on the Intel Xeon 2680 v3 multicore platform. The results show that the proposed approach on average saves 15% power, reduces 3° $c$ average temperature, improves task schedulability, and avoids 100% thermal emergencies compared to the state-of-the-art technique.