Análise comparativa de benchmarks para banco de dados NoSQL orientados a grafos de propriedades

Abstract

The new era of data economics, based on data sets with enormous volume, wide variety of formats, and rapid speed in data production and utilization, has brought the need for new data management structures and methods. Then come Not Only SQL (NoSQL) databases, which provide a new way of storing and retrieving data with features that go beyond relational database management systems. NoSQL databases are divided into four major models: key -value, column-oriented, document-oriented and graph-oriented. They are able to scalably handle the storage and processing of gigantic volumes of data with a flexible format, where manipulations are not exclusively performed through the SQL language. These data sets, many of which, modeled as large graphs, to - a challenge for industry and academia, which has been increasingly committed to research and innovation in the area. Thus, there is a natural increase in demand for performance analysis systems for Graph Oriented NoSQL Databases. This implies the need for new benchmarks capable of testing these new technologies and that guide users to identify tools that best adapt to their applications. Many studies approach the proposal of benchmarks for an analysis of Graph Oriented NoSQL Database systems. However, no study studies the comparison of such benchmarks. Thus, this Master's research work aimed to carry out a comparative analysis of benchmarks for performance evaluation of Graph Oriented NoSQL Databases, with specific emphasis on the property graph model, and thus highlight as main differences between these, in addition to identifying strengths and limitations of each benchmark. Thus, it becomes possible to identify the most appropriate benchmark to analyze the performance of Property Graph-Oriented NoSQL Database systems according to a set of requirements of a specific domain. Application. Among the several existing benchmarks, it was investigated to compare the LDBC SNB, XGDBench, HPC-SGAB, TGDB and Cyclone benchmarks. In general terms, the LDBC-SNB benchmark stood out in relation to the other benchmarks, especially when it comes to query types and performance measures, and because of its tester support for several graph-oriented NoSQL DBMS's.