Dynamic Resource Prices in a Combinatorial Grid System

Our paper presents an agent-based simulation environment for task scheduling in distributed computer systems (grid). The scheduler enables the simultaneous allocation of bundles comprising resources such as CPU time, communication bandwidth, volatile, and non-volatile memory. The resources are allocated by an iterative combinatorial auction with proxy-bidding agents trying to acquire their desired resource allocation profiles. In order to support an efficient bidding process, the auctioneer provides resource price information to the bidding agents. Due to the complementarities and substitutionalities of the bid bundles in the proposed setting, the calculation of these prices is computationally expensive. This article proposes two different price approximation mechanisms: one scarcity based-scheme and a second approach using shadow prices based on the dual formulation of the relaxed linear program of the allocation problem. The stability of both allocation mechanisms is compared in the context of a closed grid system (economy) where agents buy and sell production capacity. The objective of each agent is to acquire complementary resource capacity to increase his productivity. The system's pricing and the agents' bidding behavior is evaluated based on both measures in situations of gradually increasing resource failure to test the stability of the allocation mechanism