Prof. Dr. Frank Bellosa, Andreas Merkel

Abstract: 

Scheduling policies of multitasking operating systems partition the time of processor assignment among all runnable threads. For most schedulers it is unimportant in which way a thread utilizes its assigned resources during its time of processor control. A thread's resource utilization, however, cannot only affect the thread itself, but also subsequent scheduled threads. In the case of a thread's power consumption, the power consumption can cause throttling of the thread and subsequent threads. Due to the interference, the processor allocation between the runnable threads is not fair any longer.

In this thesis, we propose a generic design to enhance general purpose schedulers to become energy-aware. The enhanced schedulers fairly partition the system's energy among threads to favor energy-efficient threads. Furthermore, they assure that a thread's power consumption does not affect another thread's execution negatively by enforcing another thread's throttling. In order to make best use of a processor's power limit, we present an energy transfer mechanism to fairly transfer energy among threads. It permits threads having power consumptions beyond a pre-defined power limit to benefit from threads having a power consumption below the limit.

Our evaluation shows that each of our examined schedulers can become energyaware, and that they assure that each thread preserves the power limit individually. Besides, our enhanced schedulers permit to partition a system's energy fairly, even in the case of energy transfers. Due to the energy transfers and the fair energy partitioning, our enhanced schedulers - limiting each thread’s power consumption - achieve a better performance than schedulers only limiting a run-queue’s and a processor's power consumption, respectively.

BibTex: 

@diplomathesis{klee08energyawarescheduling,
  author = {Christoph Klee},
  title = {Design and Analysis of Energy-Aware Scheduling Policies},
  type = {Diploma Thesis},
  address = {System Architecture Group, University of Karlsruhe, Germany},
  month = aug # "~20",
  year = 2008,
  url = {http://i30www.ira.uka.de/teaching/theses/pasttheses/}
}