TY - JOUR
T1 - Minimizing the Maximum Processor Temperature by Temperature-Aware Scheduling of Real-Time Tasks
AU - Ozceylan, Baver
AU - Haverkort, Boudewijn R.H.M.
AU - de Graaf, Maurits
AU - Gerards, Marco E. T.
N1 - Funding Information:
This work was supported by the Nederlandse Organisatie voor Wetenschappelijk (NWO) (Dutch Research Council) through the TTW Perspectief program Zero-Energy Internet-of-Things (P15-06) within Project P5.
Publisher Copyright:
© 1993-2012 IEEE.
PY - 2022/8
Y1 - 2022/8
N2 - Thermal management is gaining importance since it is a promising method for increasing the reliability and lifespan of mobile devices. Although the temperature can be decreased by reducing processor speed, one must take care not to increase the processing times too much; violations of deadline constraints must be prevented. This article focuses on the tradeoff between performance and device temperature. We first analyze this tradeoff and show how to determine the optimal lower bound for the maximum temperature for a given set of jobs with known workloads and deadlines. To do so, we use a thermal model, which describes how future decisions impact temperature dynamics. Then, we introduce a processor scheduling algorithm that computes the resource allocation that achieves this lower bound. Consequently, our algorithm finds the optimal resource allocation for the purpose of minimizing the maximum processor temperature for a set of jobs with known workloads and deadlines. Our experimental validation shows that our thermal management algorithm can achieve a reduction of up to 15 °C (42%) of the maximum temperature when the workload is high, where a previously proposed method achieved a reduction of up to 10 °C (25%). Another advantage of our method is that it decreases the variance in the temperature profile by 16% compared to previously proposed methods.
AB - Thermal management is gaining importance since it is a promising method for increasing the reliability and lifespan of mobile devices. Although the temperature can be decreased by reducing processor speed, one must take care not to increase the processing times too much; violations of deadline constraints must be prevented. This article focuses on the tradeoff between performance and device temperature. We first analyze this tradeoff and show how to determine the optimal lower bound for the maximum temperature for a given set of jobs with known workloads and deadlines. To do so, we use a thermal model, which describes how future decisions impact temperature dynamics. Then, we introduce a processor scheduling algorithm that computes the resource allocation that achieves this lower bound. Consequently, our algorithm finds the optimal resource allocation for the purpose of minimizing the maximum processor temperature for a set of jobs with known workloads and deadlines. Our experimental validation shows that our thermal management algorithm can achieve a reduction of up to 15 °C (42%) of the maximum temperature when the workload is high, where a previously proposed method achieved a reduction of up to 10 °C (25%). Another advantage of our method is that it decreases the variance in the temperature profile by 16% compared to previously proposed methods.
U2 - 10.1109/TVLSI.2022.3160601
DO - 10.1109/TVLSI.2022.3160601
M3 - Article
VL - 30
SP - 1084
EP - 1097
JO - IEEE transactions on very large scale integration (VLSI) systems
JF - IEEE transactions on very large scale integration (VLSI) systems
SN - 1063-8210
IS - 8
ER -