Efficient Resource Allocation and Consolidation with Selfish Agents: An Adaptive Auction Approach

Abstract

Through virtualization technologies, modern enterprises can build private clouds to support the daily operations of their subsidiaries and consolidate resources from them. In making these resource allocation and consolidation decisions, they want to maximize the achieved utilities and minimize the incurred costs by their subsidiaries, respectively. However, these subsidiaries very often operate autonomously and have private information about job characteristics and energy costs. Due to this information asymmetry, they might be motivated to behave in their own best interests rather than that of the enterprise. To solve these principal-agent problems, we design a tunable auction under which subsidiaries submit bids and resources are allocated in proportion to their bids. We show that the induced competition game obtains a unique Nash equilibrium, under which hidden characteristics of subsidiaries can be revealed. By using variational inequality techniques, we derive the dynamics of the Nash equilibrium as a function of the auction parameters. We design a feedback control mechanism to dynamically adjust the tunable auction parameters based on observable information such as the bids and the resulting allocations. We prove that our adaptive auction converges to the optimal Nash equilibrium under which the aggregate utility of an enterprise is maximized.