Abstract

For decades markets have been proposed for allocating computer resources in distributed systems like the Clusters or Grids. Nevertheless, none of these approaches has made it into practice. The reasons for the adoption failure of markets are manifold. One reason that has been hitherto rarely discussed is the bidding process. Theoretic approaches assume that the bidders know how to derive their bids exactly. This does not only include the specific computer resource, which is needed at some time in the future, but also the price. This assumption simplifies reality and can thus not contribute to the development of markets in Grid. What is needed to establish a prospering market for Grid resources are rules how to conduct the bidding. Since demand and supply are extremely dynamic in computing resources, manual bidding is too slow to accommodate abrupt shifts in demand or supply. This paper introduces a policy based autonomous agent approach for automated bidding. By means of the policies resource providers and consumers can specify the way how they trade Grid resources (e.g., resource isolation definitions, security specifications).

Download

Share

COinS