DocumentCode :
2262778
Title :
DTS: Using Dynamic Time-Slice Scaling to Address the OS Problem Incurred by DVFS
Author :
Jia, Gangyong ; Gao, Xuhong ; Li, Xi ; Wang, Chao ; Zhou, Xuehai
Author_Institution :
Dept. of Comput. Sci. & Technol., Univ. of Sci. & Technol. of China (USTC), Hefei, China
fYear :
2012
fDate :
24-28 Sept. 2012
Firstpage :
65
Lastpage :
72
Abstract :
Dynamic voltage and frequency scaling (DVFS) has been the most useful technology to reduce power consumption, but it causes unpredictable program performance decreasing and unfair performance sharing among threads, which may render performance analysis, optimization, and isolation extremely difficult and lead to thread starvation and priority inversion. This paper firstly proposes an OS scheduler based on dynamic time-slice scaling (DTS) to address the OS problem incurred by DVFS. The DTS scheduler dynamically allocates each thread with a time-slice according to threads´ behavior characteristic and processor´s frequency. There are two main contributions: 1) this paper is the first work to analyze and address the OS problems incurred by DVFS, 2) we firstly allocate threads´ time-slice according to threads´ behavior characteristic and processor´s frequency. Experimental results demonstrate our DTS scheduler can well solve most OS problems incurred by DVFS with little cost, and at the same time DTS improves system performance.
Keywords :
microprocessor chips; optimisation; performance evaluation; power aware computing; power consumption; scheduling; DTS scheduler; DVFS; OS problem; OS scheduler; dynamic time-slice scaling; dynamic voltage and frequency scaling; optimization; performance analysis; power consumption; unpredictable program performance; Degradation; Dynamic scheduling; Equations; Instruction sets; System performance; Time frequency analysis; DVFS; Dynamic time-slice scaling; threads´ behavior characteristic; unfair sharing; unpredictable performance decreasing;
fLanguage :
English
Publisher :
ieee
Conference_Titel :
Cluster Computing Workshops (CLUSTER WORKSHOPS), 2012 IEEE International Conference on
Conference_Location :
Beijing
Print_ISBN :
978-1-4673-2893-7
Type :
conf
DOI :
10.1109/ClusterW.2012.12
Filename :
6355848
Link To Document :
بازگشت