New Data Placement Strategy in the HADOOP Framework

Abstract

Today, the data quantities generated and exchanged between information systems continues to increase. Storing and exploiting such quantities require can't be done without bigdata systems with mechanisms capable of meeting technological challenges commonly grouped under the four Vs (Volume, Velocity, Variety and Veracity). These technologies include mainly the Distributed File System (DFS). Like Hadoop, which is based on HDFS, the main Big Data systems use a data distributed storage where a subsystem is responsible for subdividing data (data striping) and replicating it on a network of nodes called Grid. In the typical case of Hadoop, a Grid generally consists of many nodes, grouped in multiple Racks. The logic of distributing the stored data through the Grid respects a simple strategy that guarantees the durability of the data and a certain speed of writing. This strategy does not take into consideration neither the technical characteristics of nodes, nor the number of requests on the data, which means a considerable loss in processing capacity of the grid. In this work we proposed a new placement strategy based on exploitation analysis of new information integrated into the HDFS metadata model. A significant 20% improvement in overall processing time was reached through the simulations we conducted on Hadoop.