Jämförelse mellan graf- och relationsdatabas : En studie av prestanda vid sökning av kortaste vägen mellan två givna platser i ett rälsbundet nätverk

Sammanfattning: Traditional relational databases store data in tabular form and have existed for several decades. The new requirements for data such as high availability and scalability have led to an increase in NoSQL databases in popularity. NoSQL databases meet these requirements as they use other methods for handling and storing data, for example document databases and graph databases are two of these variants. This study examined the difference in performance between the SQL Server 19 relational database and the Neo4j graph database. An experiment with the hypothesis: "Graph databases have faster response times compared to relational databases when retrieving the shortest route between two specified locations" was performed by executing a function on a dataset provided by the study's partner the Swedish Transport Administration. The data set represents Sweden's railway network and consists of 1320 places and 2788 associated connections. The function searched for the shortest route between two locations for four selected sections in each database architecture. The observed and analyzed response times show that Neo4j has an average response time that is 50 times faster than SQL Server 19, which verifies the hypothesis. The response times from the two databases were also tested with a Wilcoxon test which showed that the median response times differ from each other at a 1 % significance level. In addition, the results show that the average response time for SQL Server 19 will increase more than Neo4j as more sites and connections become involved in the search. Relational databases have slower response times than graph databases as they use join statements to find current relationships between its tables, which means that they must search all the data to find the shortest path between two places. Unlike relational databases, graph databases only use relationships directly connected to the current node where the algorithm is located, which means that response times are shorter.