Control and Optimization of the SRPT Service Policy by Frequency Scaling

• Published in: Quantitative Evaluation of Systems Vol. 11024; pp. 257 - 272
• Main Authors: Marin, Andrea, Mitrani, Isi, Elahi, Maryam, Williamson, Carey
• Format: Book Chapter
• Language: English
• Published: Switzerland Springer International Publishing AG 01.01.2018; Springer International Publishing
• Series: Lecture Notes in Computer Science
• ISBN: 3319991531; 9783319991535
• ISSN: 0302-9743; 1611-3349
• DOI: 10.1007/978-3-319-99154-2_16

In this paper, we study a system where the speed of a processor depends on the current number of jobs. We propose a queueing model in which jobs consist of a variable number of tasks, and priority is given to the job with the fewest remaining tasks. The number of processor frequency levels determines the dimensionality of the queueing process. The objective is to evaluate the trade-offs between holding cost and energy cost when setting the processor frequency. We obtain exact results for two and three frequency levels, and accurate approximations that can be further generalized. Numerical and simulation results show the high accuracy of the approximate solutions that we propose. Our experiments suggest that a parsimonius model with only two frequency levels is sufficient, since more elaborate models provide negligible improvements when optimizing the system.