RSMOA: A revenue and social welfare maximizing online auction for dynamic cloud resource provisioning

We study online cloud resource auctions where users can arrive anytime and bid for heterogeneous types of virtual machines (VMs) assembled and provisioned on the fly. The proposed auction mechanism RSMOA, to the authors' knowledge, represents the first truthful online mechanism that timely responds to incoming users' demands and makes dynamic resource provisioning and allocation decisions, while guaranteeing efficiency in both the provider's revenue and system social welfare. RSMOA consists of two components: (1) an online mechanism that computes resource allocation and users' payments based on a global, non-decreasing pricing curve, and guarantees truthfulness; (2) a judiciously designed pricing curve, which is derived from a threat-based strategy and guarantees a competitive ratio O(ln(p)) in both system social welfare and the provider's revenue, as compared to the celebrated offline Vickrey-Clarke-Groves (VCG) auction. Here p is the ratio between the upper and lower bounds of users' marginal valuation of a type of resource. The efficacy of RSMOA is validated through extensive theoretical analysis and trace-driven simulation studies.