Competitive algorithms for the weighted server problem

Abstract

The authors deal with a generalization of the k-server problem, in which the servers are unequal. In the weighted server model each of the servers is assigned a positive weight. The cost associated with moving a server equals the product of the distance traversed and the server weight. A weighted k-server algorithm is called competitive if the competitive ratio depends only upon the number of servers. (i.e., the competitive ratio is independent of the weights associated with the servers and the number of points in the metric space). For the uniform metric space, they give super exponential competitive algorithms for any set of weights. If the servers have one of two possible weights, they give deterministic exponential competitive algorithms and randomized polynomial competitive algorithms. They use the MIN operator for both algorithms. One can model the problem of storage management for RAM and E/sup 2/PROM type memories as a weighted server problem with two weights on the uniform metric space.<>