Performance Analysis of NoSQL Databases with Large Volumes of Open Educational Data

Non-Relational Databases, also known as NoSQL (Not Only Structured Query Language), emerged in the face of new requirements of Web 2.0 computer applications. Relational databases, although consolidated as a data storage and manipulation model for decades, began to face performance limitations when dealing with large volumes of data. NoSQL databases have flexible data structure, and when associated with distributed computing provide a good scalability, being indicated in the Big Data scenario. In this context, this work evaluates the performance of three NoSQL databases, in order to verify their performance in large volumes of educational data. The experiments were performed with school census data, available in the repository of the Ansio Teixeira National Institute for Educational Studies and Research (INEP) in Brazil. For this case of study, the following databases were adopted: DynamoDB (whose data model is key-value oriented), MongoDB (whose data model is document-oriented), and Cassandra (whose data model is column-oriented). Therefore, among the investigated databases, MongoDB was more efficient, presenting lower processing times in the operations of inserts/loads, queries, updates, and removals of basic educational data.