Resource Scheduling Algorithm for Multi-core System Chip with Dark Silicon
Sun Aolin1,2), Xu Qi1,2), and Chen Song1)*
1) (Department of Electrical Science and Technology, University of Science and Technology of China, Hefei 230000) 2) (Micro-/Nano-Electronic System Integration Center, University of Science and Technology of China, Hefei 230000)
With the increasing chip integration, the power density has been rising and results in on-chip thermal emergency. Recently, the dark silicon designs are proposed to avoid the hotspot, where all the fractions of chip resources cannot be simultaneously powered on. Consequently, we need reasonably determine which fraction should be powered on to ensure the throughput requirements. In this paper, we proposed a resource scheduling method for the multi-core system with dark silicon design. First of all, given a set of applications, we extracted the throughput and power consumption of each application under all the possible processor number and frequency configuration, which are constrained by the system resources and power budget, The resource configuration problem was solved by a dynamic programming-based method. Secondly, with the configuration, a simulated annealing based method is used to map the application and determine the distribution of dark cores and active cores on the system with minimization of the thermal costs and communication costs. Finally, according to the feedback of whether there exist hotspots in the system, a loop based thermal design power adaption method was used to obtain maximum power budget and avoid temperature violation. Experimental results show that, in comparison with chess mapping approach, the maximum temperature can be decreased about 3% at best. Additionally, we obtain 12% gain in performance when compared with power down hotspot adaption.