Efficient Task-Resource Matchmaking Using Self-adaptive Combinatorial Auction

Abstract

In loosely coupled distributed computing systems, one of the major duties performed by the scheduler is to efficiently manage the allocation of computational tasks to computing resources. Such matchmaking services become difficult to implement when resources belong to different administrative domains, each of which has unique and diverse valuation for task bundles. In order to cope with the heterogeneity, we introduce a novel combinatorial auction approach that solves the task-resource matchmaking problem in a utility computing environment. This auction based approach is characterized as “self-adaptive” in two senses. First, efficient allocation is achieved through adaptive adjustment of task pricing towards the market equilibrium point. Second, payment accounting is adaptive to the changing auction states at various stages that discourages strategic bidding from egocentric bidders. The objective of the research presented in this chapter is to examine the applicability of the combinatorial auction based approaches in utility computing, and to develop efficient task allocation schemes using the self-adaptive auction. Through simulations, we show that the proposed combinatorial auction approach optimizes allocative efficiency for task-resource matchmaking when valuation functions are concave, and achieves incentive compatibility once the auction process finalizes.