Research Of High Availability Replica Management And Performance Optimization In Data Grid

Data grid has resolved the puzzle that it is difficult for traditional data management systems to access, transfer and analyze the large scale distributed data. And it has boosted the development of scientific research and engineering practice extremely. Replica management was introduced into data grid for enhancing data availability, reducing network flow and improving data access performance. Nevertheless, because of the characteristics of highly dynamic, heterogeneity and large scale, data grid systems are fail to achieve high data availability and performance optimization. How to build an replica management mechanism according to the characteristics of data grid, to enhance data availability and improve data access performance, which is one of the host spot issues in literature.Based on comparison research , this dissertation summarized the replica management requirements of data grid , and constructed a dynamic replica management service model. Moreover, the author presented the corresponding algorithms, such as adaptive replica creating strategies, dynamic equilibrium replica location algorithms, replica selection algorithms based on fuzzy-grey prediction and dynamic asynchronous consistency maintenance algorithms. The main contents are as follows.â‘ According to the characteristics of data grid, the author analyzed replica data high availability requirements and performance optimization requirements respectively. Based on it, the dynamic replica management service model that it considered data availability measurement and performance optimization strategy lay was built.â‘¡Aimed at the problems of replica creating in data gird. The author chose Markov model to confirm replica redundance accurately for guaranteeing data availability through considering the influence factor of data consistency. Moreover, the dissertation presented a suit of cost shared replica creating algorithms. cost shared mechanism was established to encourage autonomous nodes creating replica, and these nodes in favor of global performance optimization. Then, the performance optimization algorithm was analyzed theoretically. At last, simulation results demonstrated the correctness and effectiveness of the algorithm.â‘¢Aimed at the problem of replica location in data grid, a dynamic equilibrium replica location algorithm was proposed on the basis of modified ant algorithm. It can self-adapt node to dynamic join or quit. With replica locate accurately as prior condition, a strategy of replica selection based on grey prediction is presented. It is not constrain to prediction samples. Through fuzzy controller compensate prediction value, it have got prediction values accurately. In the end, the author confirmed the self-study factor of fuzzy controller, and the correctness and effectiveness of the algorithm was demonstrated by simulation.â‘£Aimed at the problems of keeping replica data consistency in data grid, a keeping data consistency algorithm is presented. Combined dynamic voting mechanism, it satisfy low online probability. It improved systems performance and enhanced data availability, by means of decrease nodes that it is concerned with read and write operation. Furthermore, keeping consistency algorithm reduced cost of communication to enlarge expansibility. Its correctness is proofed from global ordered and read consistency. At last, simulation results demonstrated the correctness and effectiveness of the algorithm.To sum up, according to the replica management requirements of data grid, this dissertation proposed a suit of dynamic replica management service model, as well as a series of corresponding algorithms. It is a preferable schame for enhancing data availability and performance optimization. By means of theoretic analysis and simulation, it can be clearly concluded that the model and the algorithms are correct and effective, which can be used in data grid.