Energy efficient load balancing in web server clusters

Misikir Gebrehiwot, Samuli Aalto, Pasi Lassila

Research output: Chapter in Book/Report/Conference proceedingConference contributionScientificpeer-review

5 Citations (Scopus)

Abstract

Data centers consume substantial amounts of energy. Server clusters, being an integral part of data centers, take the largest share. Load balancing policies in server clusters have historically been optimized for performance or simplicity. However, the choice of load balancing policy, together with the power provisioning policy, can greatly impact energy efficiency of a server cluster. In this paper we consider the two widely deployed load balancing policies: the Round-Robin (RR) and Join the Shortest Queue (JSQ) policies, and compare their performance and energy consumption to a simple modification of JSQ (Pack-JSQ). Using mean response time and mean power consumption as metrics, we study these three policies via simulation runs by using publicly available Wikipedia request traces. We show that Pack-JSQ gives superior energy efficiency without compromising performance when deployed with a simple power provisioning policy that controls the number of running servers based on request arrival rate information.
Original languageEnglish
Title of host publicationProceedings of the 29th International Teletraffic Congress (ITC 29)
PublisherIEEE
Pages13-18
Number of pages6
ISBN (Electronic)978-0-9883045-3-6
DOIs
Publication statusPublished - 12 Oct 2017
MoE publication typeA4 Article in a conference publication
EventInternational Teletraffic Congress - Genoa, Italy
Duration: 4 Sep 20178 Sep 2017
Conference number: 29
http://itc29.org/

Conference

ConferenceInternational Teletraffic Congress
Abbreviated titleITC
CountryItaly
CityGenoa
Period04/09/201708/09/2017
Internet address

Keywords

  • energy-aware load balancing
  • server cluster
  • green cluster

Fingerprint Dive into the research topics of 'Energy efficient load balancing in web server clusters'. Together they form a unique fingerprint.

Cite this