The Research Of Adaptive Bidding Strategy For Agents In Dynamic Cloud Markets

Cloud computing has gradually become a public resources like water, electricity etc with the rapid development recently. As a result, how to distribute cloud resources on the cloud platform which is the schedule center of cloud computing has become the focus of computer science. Among diverse models of distribute cloud resources, the auction model turns out to become more and more common. Not only because auction model requires the least global information, but also it’s easier to be realized. Currently, more researches prefer mechanism design, instead of individual bidding strategy in cloud market. However, the value of bid price directly affects the profit the bidder can obtain. Besides, every decision in cloud should be made instantly due to the dynamic of cloud market. In order to reduce the manual intervention and maximize bidder’s profits as well as be adaptive to the dynamic change in cloud market, designing adaptive bidding strategy for Agent in cloud market has become an urgent and extremely important issue.The article performs following researches by designing adaptive bidding strategy for agents dynamic cloud market:1. Establish the general dynamic cloud market model and group-buying dynamic cloud market model.We have established the general dynamic cloud market model. This model is based on first-price sealed-bid single combinatorial auctions. In combinatorial auctions, winner determination problem is an NP hard problem. By taking the complexities as well as fairness into consideration, the model proposes an approximated WDP distribution mechanism using sort algorithm. Unlike the ordinary dynamic cloud market model, the group-buying dynamic cloud market model would have price-curves which can get more discount for more quantity demanded of virtue machines in cloud. The above model is also based on first-price sealed-bid single combinatorial auctions, with a chief agent between user-agents and cloud service provider. We also introduce the approximated WDP distribution mechanism as well as discount pricing strategy. These models above are all imperfect information markets.2. Design two adaptive bidding strategies for agent in general dynamic cloud market.We design the adaptive bidding strategy for agents as two different situations in general dynamic cloud market. The first focus on how agent perceive the dynamic in cloud market and submit the appropriate bid price. It was designed based on reinforcement learning and takes advantages of the agent’s previous experience to predict the current supply/demand ratio to adjust bid price for the most profits. The second, by avoiding the Prisoner’s Dilemma in agents competition, Q-strategy based on reinforcement learning as well as ε-greedy can also keep agents to monitor the environment of dynamic cloud market and stay away from auction frenzy.3. Design adaptive bidding strategy for agents in general group-buying dynamic cloud market.In general group-buying dynamic cloud market, at first we analyze the benefits for providers and agents because of group-buying as an incentive economic means to stimulate trades. Then we prove that it may exist the Nash equilibrium solution under this model. With the help of reinforcement learning, we design adaptive bidding strategy for agents.