Research On Replica Selection Strategy Based On Combination Algorithm In Data Grid

Grid computing, as a new type of large-scale distributed computing, has received a wide publicity in the academic community and industry community from its proposed up till now. With the continuous development of grid technology, grid has been gradually developed and broken down into a variety of grid patterns, such as computational grid, data grid, service grid as well as equipment grid. In data grid which is dealing with large data, it is not feasible for all users to access a single instance of data in terms of performance, based on reduction of bandwidth consumption and increasing data access speed and other considerations, an effective way to solve the above problem is to copy the data set as replicas of the data set and put them in different sites. How does the system select the best replica with minimum cost is an urgent issue, that is, replica selection in data grid.In this paper, the research is about replica selection strategies in data grid. The main work is as follows:First, existing replica selection strategies based on economic models and swarm intelligence methods are discussed, especially based on swarm intelligence methods, including replica selection strategies based on simulated annealing algorithm, genetic algorithm and ant algorithm.Second, after researching replica selection strategies based on genetic algorithm and ant algorithm and considering their advantages and disadvantages, a new replica selection strategy based on combination algorithm is proposed. The fitness function of genetic algorithm is used to determine pheromone distribution in ant algorithm, and then ant algorithm is used to select the best replica, so the new algorithm solves the problem of low efficiency of genetic algorithm and lack of pheromone in the beginning of ant algorithm.Third, in order to assess the performance of new replica selection strategy, grid simulation software OptorSim is selected for simulation. Three replica optimization algorithms are implemented by expanding the simulator OptorSim, including adding ant algorithm, genetic algorithm and combination algorithm, and simulation results are displayed graphically. Through analysis and comparison of simulation results, the evaluation demonstrates that the new algorithm can be effective in reducing the operational running time and improve the speed of replica selection in the whole grid environment and in a single site, so it improves overall performance.The last part summarizes all the research work and gives some prospect to the further work.