Efficient Big Data Modelling and Organization for Hadoop Hive-Based Data Warehouses

Abstract

The amount of data has increased exponentially as a consequence of the availability of new data sources and the advances in data collection and storage. This data explosion was accompanied by the popularization of the Big Data term, addressing large volumes of data, with several degrees of complexity, often without structure and organization, which cannot be processed or analyzed using traditional processes or tools. Moving towards Big Data Warehouses (BDWs) brings new problems and implies the adoption of new logical data models and tools to query them. Hive is a DW system for Big Data contexts that organizes the data into tables, partitions and buckets. Several studies have been conducted to understand ways of optimizing its performance in data storage and processing, but few of them explore whether the way data is structured has any influence on how quickly Hive responds to queries. This paper investigates the role of data organization and modelling in the processing times of BDWs implemented in Hive, benchmarking multidimensional star schemas and fully denormalized tables with different Scale Factors (SFs), and analyzing the impact of adequate data partitioning in these two data modelling strategies.