The Research On Key Technologies Of Replica Strategy Based On Cloud Storage

Cloud storage is a novel model for internet applications.Via virtualization,distributed computing technology,it can provide both highly scalable data storage and business access services for users,meanwhile it is able to meet the requirement of big data for storing resources.However,with the widely applications of the cloud storage,its application environment becomes more complex,it puts forward higher requirements for storage efficiency and data reliability.This thesis studies the key technologies of replica strategy in cloud storage,and proposes three replica algorithms:a dynamic adaptive replica placement algorithm,an online redundant replica removal algorithm and multi-replica data reliability algorithm,aiming to achieve to high load balance,low storage consumption and ensure data reliability.The main contents of this paper are summarized as follows:1.Research on replica placement in cloud storage.To deal with the problem of overload aggregation caused by nodes growing too fast under the high load cloud storage,a decentralized self-adaptive dynamic replica placement algorithm is proposed in this thesis.This method obtains optimal time according to node's overheating similarity,which can reduce the probability of node overload.Meanwhile,it applies the manner of decentralized self-adaptive to select the optimal replica placement node from its neighbor nodes,relieving the aggregate effect of being overloaded.Extensive experiments demonstrate that the average load of nodes is reduced by nearly 10%,and the average access delay of data is reduced by 15%?20%than other similar algorithms.2.Research on replica removal in cloud storage.In order to improve the reasonableness of the decision of duplicate removal,a redundant duplicate removal algorithm based on predictive evaluation mechanism is proposed.By taking into account the node load,data access and node location,this approach exploits fuzzy cluster analysis method to select the optimal deletion replica as candidate deletion replica form replica set before deleting a replica.It evaluates the impact based on the optimal deletion replica's historical visiting information and the capacity of node,deciding whether this replica can be deleted.Extensive experiments show that it can remove about 40%replicas on average,and obtains superior performances in access latency,better load balance than other similar algorithms.3.Research on the relationship among data reliability,replica placement and replica removal.To balance among the data reliability,replica placement and replica removal,a multi-replica reliability strategy is proposed in this thesis,aming to improve data access efficiency by using fewer replicas while meeting the data reliability requirement.It uses the method of "multiple times replica placement" and "redundant replica deletion" to achieve the goal.When a file stores on cloud,this strategy adopts the centralized manner to create the minimum replicas that meet the storage expectations according to file's storage expectations,and places the replicas under the guidance of global information.To respond to the time-varying dynamic cloud in time,it uses the manner of decentralized self-adaptive to dynamically create fewer replicas,and selects optimal node as placement nodes to improve load balance.Meanwhile,this approach uses a method of periodicity detection to ensure data reliability.In addition,it uses verification evaluation method to remove redundant replicas and further reduce storage consumption.Extensive experiments illustrate that the average load of nodes is reduced by about 10%,and the storage utilization can be improved by about 50%.Meanwhile,compared with the initial expectations of data,the data reliability can be increased 8%.