Strategic Bidding for Cloud Resources under Dynamic Pricing Schemes

摘要

Cloud computing offers computing and storage services which can be dynamically developed, composed and deployed on virtualized infrastructure. Cloud providers holding excess spare capacity, incentivize customers to purchase it by selling them in a market (spot market), where the prices are derived dynamically based on supply and demand. The cloud providers allow clients to bid on this excess capacity by allocating resources to bidders while their bids exceed a intermittently changing dynamic spot price. In this paper we have used game theory to model the bidding strategies of bidders in a spot market who are attempting to procure the cloud instances, as a prisoner dilemma game. We then analyze real time data from Amazon EC2 spot market to validate this model. In a single shot prisoner dilemma game mutual defection is the Nash equilibrium. We find that a majority (approx. 85%) of bidders choose to Defect which is in-line with the single shot classical prisoner dilemma game. However considering that most bidders in a spot market are repetitive bidders, we propose a Co-operation strategy which is in-line with the Iterated Prisoner Dilemma Game.