A Performance Comparison of SQL and NoSQL Databases for Large Scale Analysis of Persistent Logs

Sammanfattning: Recently, non-relational database systems known as NoSQL have emerged as alternative platforms to store, load and analyze Big Data. Most NoSQL systems, such as MongoDB, Redis, HBase, and Cassandra sacrifice consistency for scalability which means that users may not be able to retrieve the latest changes in the data but can execute faster queries. Alternatively, these systems provide what is known as eventual data consistency. Similarly, relational database systems allow relaxed levels of consistency to obtain performance improvements. In this master thesis project previous performance and scalability benchmarking experiments are reproduced and extended to two new popular state-of-the-art NoSQL database systems: Cassandra and Redis. Additionally, a relational database system not used in previous research was tested in this project, in addition to a new release of an already-tested open source relational system. The purpose of these experiments is to extend the previous evaluation to two relational systems and two non-relational database systems regardless of their data model by measuring the time needed to load and query persistent logs under different indexing and consistency settings. The results of this research show that there is no specific type of system consistently outperforming the others but the best option can vary depending on the features of the data, the type of query and the specific system.