Partitioning and bucketing in hive-based big data warehouses

Abstract

Hive is a tool that allows the implementation of Data Warehouses for Big Data contexts, organizing data into tables, partitions and buckets. Some studies have been conducted to understand ways of optimizing the performance of data storage and processing techniques/technologies for Big Data Warehouses. However, few of these studies explore whether the way data is structured has any influence on how Hive responds to queries. Thus, this work investigates the impact of creating partitions and buckets in the processing times of Hive-based Big Data Warehouses. The results obtained with the application of different modelling and organization strategies in Hive reinforces the advantages associated to the implementation of Big Data Warehouses based on denormalized models and, also, the potential benefit of adequate partitioning that, once aligned with the filters frequently applied on data, can significantly decrease the processing times. In contrast, the use of bucketing techniques has no evidence of significant advantages.