Thermal-Aware Thread and Turbo Frequency Throttling Optimization for Parallel Applications

Abstract

The number of processing cores in multicore pro-cessors has been rising to deal with the levels performance required by modern applications. Concomitantly, the operating temperature of hardware components has become a primary concern due to economic and environmental perspectives. Hence, different software (e.g., thread throttling) and hardware (e.g., dynamic voltage and frequency scaling - DVFS) strategies have also been applied to reduce the processor temperature levels without jeopardizing the application's performance. While thread throttling strategies artificially tune the degree of thread-level parallelism of applications to improve the hardware resources utilization according to their scalability issues, turbo frequencies have been employed to speed up the execution of a given appli-cation by increasing the processor's frequencies above the base. Given that, we propose Urano. It is a thermal-aware strategy that combines thread throttling and turbo mode optimization to diminish the processor operating temperature without penalizing the performance of the application. Through the execution of twelve well-known parallel applications on a modern multicore architecture, we demonstrate that Urano decreases the peak temperature by up to 17% compared to how parallel applications are executed with minimal impact on the performance.