Research And Verification Of MongoDB Fragment Cluster Load Balancing Strategy Based On Markov Forecasting Model

In recent years, with the rapid development of Web2technology, relational database of large-scale distributed data storage technology to the traditional brought hitherto unknown challenge. The relational database in the face of high-speed access, massive data under large-scale concurrent queries, unstructured data storage and database cluster scale, increasingly stretched. Therefore, a new database--NoSQL emerge as the times require.In many NoSQL, MongoDB web application more and more favored. This paper describes the characteristics of MongoDB document oriented, focusing on the divided clusters, the default MongoDB key data sheet according to the mechanism of segmentation and based on the load balancing strategy of Chunk migration. The authors found that, the migration of Chunk cluster load balancing strategy can only guarantee the stored data is evenly distributed to all nodes based on the patch does not consider, each piece of data access node on the heat balance.Aiming at this problem, the introduction of Markov stochastic process, presents a Markov forecasting model based on load balancing strategy. First of all, the classified statistics on CRUD operating state of each Chunk fragment in the cluster, use these statistics form the corresponding homogeneous Markov chain, according to a series of characteristics of homogeneous Markov chain based on cluster load forecasting model of the balance sheet, then the prediction model will predict the next period of time the value of the load information, finally according to the predictive value of a more reasonable data transfer, prompting patch clusters to reach the level of the data accessing frequency equalization.Finally, the test data provided by the use of URP system of Yunnan University, built MongoDB sharded cluster production environment, Markov load balancing strategy and the default MongoDB equilibrium strategy of Chunk migration based on experimental results were compared based on the proposed, and analyzed the result.