Research On Replication Strategy In Cloud Environment

With the rapid development of information technology, global data storage volumes haveexperienced explosive growth in recent years. With the advantage of high scalability, faulttolerance, and high cost performance, cloud storage has become the focus of the research.Cloud storage system is typically composed of multiple data centers and storage resources. Asa platform for distributed storage, its underlying hardware is usually formed by a mount ofcheap servers. How to ensure the system’s fault tolerant and reliability has become a veryimportant issue. Replication Technology, as a mainstream technology to improve systemperformance, has resolved data failure from a single point failure by creating multiple copiesof the data and storing in different storage nodes. Meanwhile, Replication Technology helps toreduce access latency, improve the utilization rate of network bandwidth and the availability ofdata. Replication Technology can be divided into two aspects: Static Replication Strategy andDynamic Replication Strategy. At present, most systems adopt Static Replication Strategy. Thepositions and the number of replications have been confirmed before the data come into thesystem. This way is simple but lack of flexibility, and doesn’t take into account the userrequirement, the nature of storage node and the influence of factors that change thesurrounding environment. It can cause unreasonable utilization of storage resources to acertain extent.In order to take more reasonable use of the cloud storage resources, abundant the researchof Replication Technology, we mainly study on the research of data replication strategy andpropose a dynamic replication strategy based on relevant failure. In this paper, the researchincludes the following components:(1) Based on the deep research of the layered architectureof the traditional center and VL2tree structure put forward by Microsoft, we build theNetwork Structure model and the Function Structure model;(2) We build the Relevant Failuremodel considering network, refrigeration, electric power, natural disasters and the distance between the storage nodes, and the Single Point Failure model considering the differentproperties of the node itself and node cascade response. On the basis of the Relevant Failuremodel and the Single Point Failure model, the reliability values of the data that is stored underthe data center with different number of copies can be calculated;(3) There are differentrequirements of reliability for different user’s data in cloud. With considering the userrequirements and the factors influenced on data reliability, i.e., the node reliability andRelevant Failure Degree, Data Replication Strategy based on Relevant Failure is provided inthis paper. The strategy comprehensively analyses the influence on the relevant failurebetween the nodes in data center, such as geographical and environmental factors. Thismethod can reduce the number of replicas as much as possible and meet the user requirementof reliability at the same time. According to the results of the experiments, our strategy fullyconsiders the relevance between the nodes and the user requirements compared with Amazon’sstatic replication strategy. Besides, the strategy has a better performance on reducing the datastorage space and increasing the utilization of storage resources efficiently.