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.

Control and optimization of the SRPT service policy by frequency scaling

Marin, Andrea
;
2018

Abstract

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.
Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
File in questo prodotto:
File Dimensione Formato  
Marin2018_Chapter_ControlAndOptimizationOfTheSRP.pdf

non disponibili

Tipologia: Versione dell'editore
Licenza: Accesso chiuso-personale
Dimensione 482.76 kB
Formato Adobe PDF
482.76 kB Adobe PDF   Visualizza/Apri
main.pdf

accesso aperto

Tipologia: Documento in Post-print
Licenza: Accesso gratuito (solo visione)
Dimensione 536.74 kB
Formato Adobe PDF
536.74 kB Adobe PDF Visualizza/Apri

I documenti in ARCA sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: http://hdl.handle.net/10278/3711312
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 7
  • ???jsp.display-item.citation.isi??? 4
social impact