Resource augmentation in load balancing

Yossi Azar, Leah Epstein, Rob Van Stee

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


We consider load balancing in the following setting. The on-line algorithm is allowed to use n machines, whereas the optimal off-line algorithm is limited to m machines, for some ffixed m < n. We show that while the greedy algorithm has a competitive ratio which decays linearly in the inverse of n=m, the best on-line algorithm has a ratio which decays exponentially in n=m. Specifically, we give an algorithm with competitive ratio of 1 + 1/2n/m(1−o(1)), and a lower bound of 1 + 1/en/m(1+o(1)) on the competitive ratio of any randomized algorithm. We also consider the preemptive case.We show an on-line algorithm with a competitive ratio of 1 + 1=en/m(1+o(1)). We show that the algorithm is optimal by proving a matching lower bound. We also consider the non-preemptive model with temporary tasks. We prove that for n = m + 1, the greedy algorithm is optimal. (It is not optimal for permanent tasks).

Original languageEnglish
Title of host publicationAlgorithm Theory - SWAT 2000 - 7th Scandinavian Workshop on Algorithm Theory, 2000, Proceedings
EditorsMagnús M. Halldórsson
PublisherSpringer Verlag
Number of pages11
ISBN (Print)3540676902, 9783540676904
StatePublished - 2000
Externally publishedYes
Event7th Scandinavian Workshop on Algorithm Theory, SWAT 2000 - Bergen, Norway
Duration: 5 Jul 20007 Jul 2000

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349


Conference7th Scandinavian Workshop on Algorithm Theory, SWAT 2000

Bibliographical note

Publisher Copyright:
© Springer-Verlag Berlin Heidelberg 2000.

ASJC Scopus subject areas

  • Theoretical Computer Science
  • General Computer Science


Dive into the research topics of 'Resource augmentation in load balancing'. Together they form a unique fingerprint.

Cite this