Comparison of relational and columnar database systems

In this document two different types (relational and columnar) of databases will be compared by analysing differences between various data storage realisations in each type of selected databases, looking for the best data storage choice when saving particular type of data. The objective of this work is to overcome the performance issues when working with a big set of analytical data by selecting the best possible data storage format and database which implements selected structure the best. During the process the test was carried out. A big set of data was prepared for various agregation operations (such as calculation of averages). Results have showed that analysing(reading data from database) is done better by columnar type of database, especially when working with bigger sets of data. However, relational databases are easier to set up and perform almost the same or even better when working with little sets of data. The whole test was performed using Ruby programming language (no framework was used due to possibility of unwanted overhead which might make the results less accurate) and three databases: “Cassandra” as row-oriented columnar database, Druid as completely columnar database which is configured to work best when working with analytical data and “Postgresql” as relational database. Three tests were made in total, each of them had different data set size with the purpose of finding the point, when switching from one database to another is the best option