| The grid environment allows users to access cross-regional distributed computing resources without knowing it. The grid resources scheduling mechanism must coordinate participants in grid and attract resources providers and users involved in grid resources interaction. Auction model with little global information has the distributed structure and respect the autonomy of resource providers, so auction model is a very good way to solve the resource allocation problem.This paper proposes a multi-attribute auction mechanism for grid resources scheduling which describe users’ need in four method, including the single resource needs, the combination resource needs, the task needs and and multi-attribute demand. The grey theory and support vector machine regression method predict the unit-price in current auction, and system calculated the split proportion of combination offer and all offers can bid together. Resources matching mechanism use multiple attribute decision making method, price priority strategy and demand for resources priority strategy matching respectively in different situation.System receive the result of each single resource to estimate whether the combination need success or not. The final deal price is determined by the market structure parameters. For the users and resource provider, the risk attitude parameters, the failure of the emotional parameters and joint bid strategies are effective in resources bid. Finally the access control mechanism and the reputation model ensure the good safety performance in grid.In order to check the feasibility and the validity of the designed mechanism, simulated experiments are made to build the multi-attribute auction mechanism with Simjava toolkit compare to the DAM algorithm Simulation results have shown that the mechanism hava a good performance in job success rate, resource utilization rate and load balancing, and in fairness deviation of profit index also need to improve. |
PDF Full Text Request