Implementation and Evaluation of a Software-based Power Virtualization Technique

Open Access
Author:
Bily, Paul Lawrence
Graduate Program:
Computer Science and Engineering
Degree:
Master of Science
Document Type:
Master Thesis
Date of Defense:
March 18, 2013
Committee Members:
  • Bhuvan Urgaonkar, Thesis Advisor
Keywords:
  • power
  • energy
  • virtualization
  • virtualize
  • management
  • data center
  • datacenter
  • DVFS
  • UPS
  • PDU
  • hierarchy
  • lottery scheduling
Abstract:
As data centers grow in size and popularity, managing utilization of resources like power becomes increasingly important due to rising costs. Effective power management remains challenging due to the dynamic needs of applications across different executions and hardware configurations resulting in occasional peaks in usage that make efficient power allocation difficult. Thus, a software-based virtualization technique is presented that helps coordinate current power techniques, such as DVFS, battery hierarchies, and overbooking capacity, to more effectively allocate and manage power usage across a data center. In particular, it considers how to fairly reallocate power during a power emergency. An implementation is presented and tested via simulation to examine its efficacy. This demonstrates how coordinating power usage in a data center can robustly virtualize power and allow for more aggressive overbooking while maintaining performance and fairness considerations.