Analysis of Dynamic Data Placement Strategy for Heterogeneous Hadoop Cluster

ABSTRACT MapReduce has become a very important distributed process model for large scale data-intensive applications like Web Data and Data Mining. Hadoop is an open source implementation of MapReduce is wide used for large data processing which requires low time response. This paper, address the matter of approach to place data across nodes in an exceedingly way that every node contains a balanced processing load. Given an Data intensive application running on a Hadoop MapReduce cluster, our Data placement theme adaptively balances the number of knowledge hold on in every node to realize improved data-processing performance. Experimental results show that our Data placement strategy will forever improve the MapReduce performance by rebalancing information across nodes before playacting a data intensive application in an exceedingly heterogeneous Hadoop cluster. It is necessary for data placement algorithms to partition the input and intermediate data supported the computing capacities of the nodes within the cluster. This Hadoop implementation assumes that each node in an exceedingly cluster has an equivalent computing capability which the tasks are data-local, which can increase further on top of and scale back MapReduce performance. This paper proposes a dynamic data placement algorithm to resolve the unbalanced node employment downside. The planned technique will dynamically adapt and balance data hold on in every node supported the computing capability of every node in an exceedingly heterogeneous Hadoop cluster. The planned algorithm will scale back data transfer time to realize improved Hadoop performance. The experimental results show that the dynamic information placement policy will decrease the time of execution and improve Hadoop performance in an exceedingly heterogeneous cluster.