Efficient Bidding for Virtual Machine Instances in Clouds

Combinatorial auctions are efficient mechanisms for allocating Virtual Machine (VM)instances to cloud computing users. Despite the fact that in general these mechanisms lead to higher revenues than the currently employed fixed-price mechanisms, the cloud computing providers do not employ them to allocate their resources. One of the main reasons is the complexity faced by the users when determining the bid (i.e., the bundle of VM instances and the bid value). We address this issue by developing an efficient bidding strategy for the users requesting VM instances. We design new metrics for evaluating bundles of VM instances based on the characteristics of the computing tasks. These metrics allow us to determine the valuation of the requested bundles and to design algorithms for selecting the best bundles to bid for. We perform simulation experiments to evaluate the proposed strategy in a simulated cloud environment.