MODELAGEM DE DADOS EM DATA WAREHOUSES MODERNOS: COMPARAÇÃO ENTRE MODELOS NO CONTEXTO NAS NOVAS PLATAFORMAS ANALÍTICAS

Abstract

After the second half of the 1980s, companies only stored data from their transactional systems. However, naturally, the need arose to obtain metrics based on data that could support decision makers in their management activities. As a result, several types of Decision Support Systems were developed, such as Data Warehouses. This is a technology in which data is extracted from transactional systems, subsequently transformed and loaded into a database. In this way, end users were able to perform analyses from multiple perspectives through a single, integrated source of data. This was a successful model for a long time, until the 2000s saw an exponential growth in the amount and variety of data generated by organizations. This spurred the development of technologies for distributed storage and processing such as Hadoop, and then cloud computing platforms such as Azure, AWS, and Google Cloud. This new context of analytical environments has brought about important changes, such as a significant decrease in data storage costs and the decoupling of processing and storage. In view of this, it is natural to ask questions such as: do traditional data models like Star Schema still make sense nowadays or is the best option to embrace bolder proposals like One Big Table? When investigating what data professionals are thinking about the subject, one realizes that there is no consensus around the topic. This is because each specific case presents its own peculiarities, so that no model will meet the needs of all situations. However, despite these limitations, it is possible to achieve a balanced result between storage, maintenance, and performance by knowing the advantages and disadvantages presented by each of them.