Abstract

Encountering the increasing demand for high-performance computational resources in academic as well as commercial organisations, utility computing offers a solution by providing users with on-demand availability of requested computing services. Approaches to the fundamental issue of resource allocation include the use of technical scheduling mechanisms as well as introducing economic ideas into the allocation schemes. Technical scheduling mechanisms are often very simple (such as first-in-first-out) but suffer under the shortcoming to adequately prioritize jobs in times when demand exceeds supply. As empirical studies show, Grids (such as PlanetLab) are frequently characterized by huge excess demand for resources. This is where economic models such as markets come into play. Hitherto, market mechanisms are either (too) simple or too complex for usage in Grids.

The contribution of this paper is threefold. Firstly, a mechanism for Grids is proposed, which is still simple but geared up for use in the Grid. Secondly the mechanism is embedded in state-of-the-art Grid middleware Sun N1 Grid Engine 6. Thirdly, it is shown by means of a numerical case study that this mechanism is superior to other commonly used mechanisms.

Share

COinS