Due to the extensive use of SQL, the number of SQL-on-Hadoop systems has significantly increased, transforming Big Data Analytics in a more accessible practice and allowing users to perform ad-hoc querying and interactive analysis. Therefore, it is of upmost importance to understand these querying tools and the specific contexts in which each one of them can be used to accomplish specific analytical needs. Due to the high number of available tools, this work performs a performance evaluation, using the well-known TPC-DS benchmark, of some of the most popular Big Data Analytical tools, analyzing in more detail the behavior of Drill, Hive, HAWQ, Impala, Presto, and Spark.

In the past years, Big Data has become a hot topic across several business areas. One of the main concerns regarding this concept is how to handle the massive volume and variety of data efficiently. Due to the notorious complexity of the data associated to the Big Data concept, usually motivated by data volume, efficient querying analysis mechanisms are mandatory for data analysis purposes. Motivated by the rapidly development of tools and frameworks for Big Data, there is much discussion about querying tools and, specifically, those more appropriated for specific analytical needs. This paper explores some of the available querying tools, describing and comparing their main characteristics and architectures, crucial knowledge for selecting the more appropriate ones for inclusion in a specific Big Data analytical architecture.

Today’s system users demand fast answers when querying their own databases. Their impatience still high when waiting for the results of a query when they take more than one or two seconds to appear on the screen. However, having fast querying answers it is not the only aspect that determines the quality of a database system we are using, but also the energy consumption involved with. The development of database systems increasingly economic in terms of energy consumption has led to great technological advances in this area. Today, many of the entities that manage large data base systems pay particular attention to this issue, not only for environmental reasons but also for economic reasons, obviously. In this paper we address the issue of queries energy consumption evaluation in database systems, with particular emphasis to those that are executed in a environment of a document store. Based on the information provided by the execution of a query in MongoDB, we designed and developed a process that determines the energy consumption of queries launched in a document store, approaching different alternatives in query designing, implementation and execution.

Nos últimos anos, o termo Big Data tornou-se um tópico bastanta debatido em várias áreas de negócio. Um dos principais desafios relacionados com este conceito é como lidar com o enorme volume e variedade de dados de forma eficiente. Devido à notória complexidade e volume de dados associados ao conceito de Big Data, são necessários mecanismos de consulta eficientes para fins de análise de dados. Motivado pelo rápido desenvolvimento de ferramentas e frameworks para Big Data, há muita discussão sobre ferramentas de consulta e, mais especificamente, quais são as mais apropriadas para necessidades analíticas específica. Esta dissertação descreve e compara as principais características e arquiteturas das seguintes conhecidas ferramentas analíticas para Big Data: Drill, HAWQ, Hive, Impala, Presto e Spark. Para testar o desempenho dessas ferramentas analíticas para Big Data, descrevemos também o processo de preparação, configuração e administração de um Cluster Hadoop para que possamos instalar e utilizar essas ferramentas, tendo um ambiente capaz de avaliar seu desempenho e identificar quais cenários mais adequados à sua utilização. Para realizar esta avaliação, utilizamos os benchmarks TPC-H e TPC-DS, onde os resultados mostraram que as ferramentas de processamento em memória como HAWQ, Impala e Presto apresentam melhores resultados e desempenho em datasets de dimensão baixa e média. No entanto, as ferramentas que apresentaram tempos de execuções mais lentas, especialmente o Hive, parecem apanhar as ferramentas de melhor desempenho quando aumentamos os datasets de referência.