Using Classification Techniques to Improve Replica Selection in Data Grid

Data grid is developed to facilitate sharing data and resources located in different parts of the world. The major barrier to support fast data access in a data grid is the high latency of wide area networks and the Internet. Data replication is adopted to improve data access performance. When different sites hold replicas, there are significant benefits while selecting the best replica. In this paper, we propose a new replica selection strategy based on classification techniques. In this strategy the replica selection problem is regarded as a classification problem. The data transfer history is utilized to help predicting the best site holding the replica. The adoption of the switch mechanism of replica selection model avoids a waste of time for inaccurate classification results. In this paper, we study and simulate KNN and SVM methods for different file access patterns and compare results with the traditional replica catalog model. The results show that our replica selection model outperforms the traditional one for certain file access requests.