Design And Simulated Implementation Of Grid Resource Double Combinatorial Auction Mechanism Based On Intelligent Agents

Grid is an emerging computing platform to solve complex scientific problems, which contains a lot of heterogeneous and distributed resources belonging to different administration domain. It brings great challenges for grid resource management. Traditional system-side resource management strategies just consider system’s throughput and the total task time, without considering user-side utility, and it also cannot motivate resource owner to contribute his resources. Grid economics can effectively solve above mentioned problems, and auction model can promote competition and allocate resource better with higher economic efficiency.Grid resource double combinatorial auction mechanism based on intelligent agents is designed in this thesis, which includes grid resource double combinatorial auction protocol and intelligent agents’bidding price decision mechanism. In the former part, users and resources submit sealed bids simultaneously. Grid information server as an auctioneer takes charge of matching bids. Winner determine of combinatorial auction is an NP complete problem. Because of limitations of traditional algorithms, paddy field algorithm, championship league algorithm, bat optimization algorithm and group search optimization algorithm are introduced to solve the winner determine problem. In order to improve the satisfaction of users and resources, winners’reputation is considered besides market surplus. Reputation system based on feedback rating is designed, and reputation is directly calculated by feedback rating. Reputation fading coefficient and evaluation of users’credit are introduced to weaken influence to system brought by malicious behavior, which can enhance the reputation system’s robustness. At the latter part, bidding price decision mechanism based on back propagation neural network is designed to increase earnings of users and resources, which can reduce losses brought by acquiring inadequate knowledge. In ordor to adapt to changes of market dynamically and make scienfic decision, various fators affecting bidding price are considered. Bidder’s history bidding data is used to train the back propagation neural network.Grid resource double combinatorial auction mechanism based on intelligent agents is simulated on GridSim platform. According to the result of performance evaluation, double combinatorial auction protocol can improve market surplus and the amount of successful dealing tasks under the conditions of a little more time cost, and bidding price decision mechanism based on back propagation neural network can improve earnings of users and resources. Simulation results show that the auction mechanism designed in this thesis is practical and effective.