A Randomized Pay-as-Bid Mechanism for Grid Resource Allocation

Grid technology denotes a computing paradigm in which computing resources are shared across geographical and administrative boundaries. Market-based mechanisms are deemed promising to provide a better fit to grids' strategic and dynamic nature than classic technical schedulers. In this paper, we present a deterministic heuristic that achieves the distinct trade-off between scalability by approximation on the one hand and truthful prices for resource requests on the other hand. However, in special cases, the heuristic might generate efficiency far from optimal. At the core of this paper, we thus elaborate a randomized heuristic that might help in intercepting such worst cases. We show how this randomized heuristic can be turned to a mechanism that is truthful in expectation. However, this pricing scheme is infeasible in practice, and we thus propose the use of the simple yet promising pay-as-bid pricing rule in conjunction with a discount for high bidders in order to induce requesters to bid close to their true valuations. The evaluation of the randomized heuristic's strategic and allocative properties provides promising results.