On-line learning based dynamic thermal for multicore systems

초록

Power consumption of a high-end microprocessor very rapidly. High power consumption will lead to increase in chip temperature as well. If temperature beyond a certain level, chip operation becomes either or unreliable. Therefore various approaches for dynamic management (DTM) have been proposed. In this paper, propose a new application-oriented learning-based dynamic management (LDTM) technique for a multi-core system. repetitive executions of an application, we learn the patterns of the chip, and we control the future through DTM. When the predicted temperature rise above a threshold value, we reduce the temperature by the operation frequency of the corresponding core. implement our learning-based thermal management on an 's dual core system which is equipped with digital thermal (DTS). The dynamic frequency scaling (DFS) is to have three frequency steps on a Linux kernel. carried out experiments using Phoronix Test Suite for Linux. The peak temperature has been reduced on average 7 using our LDTM, and the overall average reduced from 72 C to 65 C.

키워드