LAUSR

The selection of cloud resources is important to users which influences their utility directly. To improve the users’ utility of purchased resources, according to the historical bidding information, we present data-driven cloud resource procurement (CRP) auctions which can help the resource buyer make an optimized selection for cloud resources. First, we design two procurement auction mechanisms with personalized reserve prices for CRP: Lazy-CRP and Eager-CRP, in which the reserve prices are set by the broker before the auction based on the broker’s knowledge about costs of cloud providers. Then, given the assumption that each cloud provider is myopic, we propose the optimal learning algorithms of personalized reserve prices for the Lazy-CRP approach, and 2-approximation learning algorithm for the Eager-CRP approach. Both of two algorithms can be performed even if the cost distribution functions of cloud providers are nonidentical. By comparing our mechanisms with Vickery–Clark–Groves CRP (VCG-CRP) in simulation, the results show that our proposed mechanism is very suitable for occasions where the number of providers is very small. According to the results of the simulation, the total utilities obtained by the data-driven procurement schemes are significantly higher than VCG-CRP when the number of cloud providers is less than four. When there are only two providers, the increased total utility of our mechanism can be up to 80% of that obtained by VCG-CRP.