TY - JOUR

T1 - A competitive 2-server algorithm

AU - Irani, Sandy

AU - Rubinfeld, Ronitt

N1 - Funding Information:
* Research supported by a Tandem Fellowship. * * Research supported by an IBM Graduate Fellowship and NSF Grant CCR 88-13632.

PY - 1991/7/31

Y1 - 1991/7/31

N2 - The K-server problem is the problem of planning the motion of K mobile servers in a metric space. We give an on-line algorithm for the 2-server problem in any metric space. The total cost of this algorithm on any sequence of requests is bounded by ten times the cost of the optimal off-line algorithm on that sequence. The rule is a modified version of the balance algorithm; it sends the server that minimizes the quality: (total distance traversed so far by that server + twice the distance of that server to the next request). This is the first provable competitive rule that can be evaluated in a constant number of arithmetic operations per request with only one variable. This contrasts with the 2-competitive 2-server algorithm in [5] which requires maintaining O(t) memory locations and O(t) time to decide which server to send, where t is the minimum of the number of points in the metric space and the number of requests. Our rule naturally generalizes to more than two servers, and we conjecture that it is also competitive in this case.

AB - The K-server problem is the problem of planning the motion of K mobile servers in a metric space. We give an on-line algorithm for the 2-server problem in any metric space. The total cost of this algorithm on any sequence of requests is bounded by ten times the cost of the optimal off-line algorithm on that sequence. The rule is a modified version of the balance algorithm; it sends the server that minimizes the quality: (total distance traversed so far by that server + twice the distance of that server to the next request). This is the first provable competitive rule that can be evaluated in a constant number of arithmetic operations per request with only one variable. This contrasts with the 2-competitive 2-server algorithm in [5] which requires maintaining O(t) memory locations and O(t) time to decide which server to send, where t is the minimum of the number of points in the metric space and the number of requests. Our rule naturally generalizes to more than two servers, and we conjecture that it is also competitive in this case.

KW - Design of algorithms

KW - amortized analysis

KW - analysis of algorithms

KW - on-line algorithms

UR - http://www.scopus.com/inward/record.url?scp=0026191154&partnerID=8YFLogxK

U2 - 10.1016/0020-0190(91)90160-J

DO - 10.1016/0020-0190(91)90160-J

M3 - ???researchoutput.researchoutputtypes.contributiontojournal.article???

AN - SCOPUS:0026191154

SN - 0020-0190

VL - 39

SP - 85

EP - 91

JO - Information Processing Letters

JF - Information Processing Letters

IS - 2

ER -