TY - JOUR
T1 - Integrating virtualization, speed scaling, and powering on/off servers in data centers for energy efficiency
AU - Gallego Arrubla, Julian A.
AU - Ko, Young Myoung
AU - Polansky, Ronny J.
AU - Pérez, Eduardo
AU - Ntaimo, Lewis
AU - Gautam, Natarajan
N1 - Funding Information:
This research was partially supported by NSF grant CMMI-0946935. The authors would like to thank the editors and the anonymous referees for their comments, which resulted in a significant improvement in the presentation of this article.
PY - 2013/10/1
Y1 - 2013/10/1
N2 - Data centers consume a phenomenal amount of energy, which can be significantly reduced by appropriately allocating resources using technologies such as virtualization, speed scaling, and powering off servers. This article proposes a unified methodology that combines these technologies under a single framework to efficiently operate data centers. In particular, a large-scale Mixed Integer Program (MIP) is formulated that prescribes optimal allocation of resources while incorporating inherent variability and uncertainty of workload experienced by the data center. However, only for small to medium-sized clients it is possible to solve the MIP using commercial optimization software packages in a reasonable time. Thus, for large-sized clients a heuristic method is developed that is effective and fast. An extensive set of numerical experiments is performed to illustrate the methodology, obtain insights on the allocation policies, evaluate the quality of the proposed heuristic, and test the validity of the assumptions made in the literature. The results show that gains of up to 40% can be obtained by using the integrated approach rather than the traditional approach where virtualization, dynamic voltage/frequency scaling, and powering off servers are done separately.
AB - Data centers consume a phenomenal amount of energy, which can be significantly reduced by appropriately allocating resources using technologies such as virtualization, speed scaling, and powering off servers. This article proposes a unified methodology that combines these technologies under a single framework to efficiently operate data centers. In particular, a large-scale Mixed Integer Program (MIP) is formulated that prescribes optimal allocation of resources while incorporating inherent variability and uncertainty of workload experienced by the data center. However, only for small to medium-sized clients it is possible to solve the MIP using commercial optimization software packages in a reasonable time. Thus, for large-sized clients a heuristic method is developed that is effective and fast. An extensive set of numerical experiments is performed to illustrate the methodology, obtain insights on the allocation policies, evaluate the quality of the proposed heuristic, and test the validity of the assumptions made in the literature. The results show that gains of up to 40% can be obtained by using the integrated approach rather than the traditional approach where virtualization, dynamic voltage/frequency scaling, and powering off servers are done separately.
KW - Data center operations
KW - energy conservation
KW - heuristic method
KW - mixed integer programming
KW - non-homogeneous systems
UR - http://www.scopus.com/inward/record.url?scp=84880205784&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=84880205784&partnerID=8YFLogxK
U2 - 10.1080/0740817X.2012.762484
DO - 10.1080/0740817X.2012.762484
M3 - Article
AN - SCOPUS:84880205784
SN - 0740-817X
VL - 45
SP - 1114
EP - 1136
JO - IIE Transactions (Institute of Industrial Engineers)
JF - IIE Transactions (Institute of Industrial Engineers)
IS - 10
ER -