Research On Dynamic Replication Strategy In Cloud Storage

With the rapid development of Cloud Computing and Big Data technology,the issues of data storage have attracted more and more attention.Users can pay for the storage service on demand.Therefore,the cloud storage with high reliability,low cost and automatic fault tolerance has become the common goal of both sides of the service.Cloud replication strategy can effectively improve the storage performance of system.Research on effective cloud replication strategy has become a hot topic in the current big data storage technology.This thesis studies the dynamic replicas strategy,which includes the optimal number of file's replicas and the dynamic layout strategy of replicas.The prototype of cloud replicas strategy visualization system is implemented.The main achievements of this thesis include as follows:(1)In order to reduce the frequency of the replicas' creation/deletion,a dynamic predicted replication strategy is proposed.This algorithm uses the single exponential smoothing to predict the amount of file's future access and tries to calculate the number of replicas based on the current amount of file's access and its future predicted amount.The experimental results proved that this algorithm can effectively reduce the access latency by adjusting the number of replicas dynamically.When the amount of access change greatly,this algorithm can decrease the frequency of the replicas' creation/deletion by predicting the future demand.Meanwhile,the number will be more reasonable when the deviation of predicted value is controlled in a certain range.(2)In order to improve the quality of cloud storage service,it is necessary to improve the multi-objectives at the same time.So a dynamic distribution strategy of replicas based on multi-objectives optimization is proposed in this thesis.Three most important performance objectives are considered for multi-objective optimization,including the access latency,load balancing and system reliability.The artificial immune algorithm is used to determine the optimal distribution of replicas.The multi-objectives optimization is considered when our strategy dynamically assigns the replicas to different storage nodes,so this strategy can achieve a high degree of reliability,and keep the load well-balanced.(3)Visualization technology is used to provide visual management and decision support for managers with figure in this thesis.Meanwhile,EChart is used to implement the visualization of above algorithms.