Research On Grid Resource Allocation And Pricing Based On The Economic Theory

Grid is a new technology which provides high performance computing, management and service for the users, by connecting the geographically distributed, heterogeneous and dynamic resources through Internet or private network. Due to the reason that the resources in Grid are geographically distributed, autonomous, and could join and leave the Grid environment at any time, the Grid resource management is regarded as a complicated issue. As there are many similarities between the Grid system and the human economic system, it is feasible and helpful to solve the problem of the Grid resource management with the aid of the economic theory.This dissertation mainly researches the resource allocation and pricing scheme in Grid resource management system. Based on the combinatorial double auction in economic theory, we carried out the researches on a series of problem, including Grid resource allocation model, resource allocation and pricing algorithm, trust mechanism, application scenarios. The work in the dissertation can be summarized as follows:With the investigation on the existing economy-based resource allocation and pricing schemes, we analysis the advantage of the combinatorial double auction theory and show how it will benefit the resource allocation in Grid. A Resource Allocation Model for the Combinatorial Double Auction (RAMCoDA) is proposed, in which the users can bid for the resource combination to buy the resources required by their tasks, while the resource providers can sell the resources they owned in the form of resource combination. The proposed scheme can remarkably improve the efficiency of the resource allocation.To complete the combinatorial double allocation scheme, a novel resource allocation and pricing algorithm, which is suitable for RAMCoDA, is designed. In the algorithm, the Grid Market Auctioneer (GMA) selects the traders according to all the participants' bids, and the bids will also be considered in the following allocation and pricing steps. By this way, all the traders in RAMCoDA can obtain the whole information of their trade objects and the prices they should pay or charge. Experiments show that the proposed scheme can complete the resource allocation and pricing efficiently, represent the incentive compatibility characteristic to some extend, achieve the stable price level, and reflect the relationship of the supply and the demand. The buyers with the higher bids and the sellers with the lower bids will get the payoff during the pricing process. Moreover, for the implementation of the resource allocation and pricing scheme, two solutions is introduced, including centralized solution and distributed solution. The component interactions and the allocation steps are described for both solutions.Based on the combinatorial double allocation proposed above, some deep researches are carried out in two items: Firstly, to further improve the efficiency of the resource allocation, the multi-round combinatorial double allocation scheme is proposed. The decision rule and the bid updating rule of the participants before they enter into the new round are also defined. Experiments show that the scheme is effective. Secondly, considering the resource combination requirement not existing in all Grid application scenes, we investigate the resource allocation scheme under the single-item multi-unit task environment. In this situation, the above resource allocation scheme reduces to the Single-item Multi-unit Double Auction (SMDA). The characteristics of SMDA are analyzed, including incentive compatibility, budget-balanced, and mechanism efficiency. An improved strategy to obtain the adjustable income of the auctioneer is also proposed. Simulation results demonstrate that, compared with the traditional Multi-unit Double Auction (MDA), SMDA can achieve a higher node trade rate and a lower efficiency loss rate, as well as guarantee the income of the auctioneer.Considering there existing the possibility that the malicious nodes exist in Grid environment and disturb the order of Grid market, we also carry out some works on trust mechanism. A simple Grid trust model is proposed to provide the trust reference while doing the resource allocation. The proposed model is combined with the combinatorial double resource allocation scheme, and the trust-bid joint allocation scheme is given. The trust-bid joint allocation scheme can consider both the trust degree and the bid of the participant node in the resource allocation process, thus preventing the malicious nodes damaging the Grid market. Moreover, classified according to the trust value type used and trust importing method, the trust-bid joint allocation scheme can be achieved in three different ways. Simulation results show that these three methods can all effectively prevent the malicious nodes entering into the market, while doing the resource allocation efficiently.Finally, to show the potential of the proposed resource allocation schemes in the actual Grid application, two scenarios, including Large File Transfer (LFT) and Mobile IPTV, are illustrated. It also shows that how to map the proposed resource allocation schemes into the two scenarios. Moreover, based on the Mobile IPTV scenario, some extended researches are carried out. A reliable scheme for IP service transmission is proposed, to ensure that the program can be sent to the mobile terminal reliably, even under the Mobile IPTV scenario with the mobile TV architecture not including IP layer.