A two-stage evaluation of user query performance for the relational model and SQL

Abstract

What are the impacts of a data model and a query language on user query performance? This is a longstanding research question about database query. Current knowledge has provided their combined effect. A data model comprises structures and operations, and could be decoupled from a query language. It is theoretically possible to identify the effects from a data model (without the query language), and the additional effects when a query language is included. An experiment was conducted to provide answers on these effects. Subject query performance with the relational model and SQL was measured at two query stages: the query translation and query writing stages. The experiment confirms literature findings about SQL query difficulties (which are all based on the query writing stage). Exploratory analysis of query difficulties show surprises. For example, operations generally perceived to be difficult (such as joins, group count and repeated relations) are not difficult at the query translation stage, i.e. the difficulties are not because of the relational model, but because of SQL. The study illustrates an approach for separating the effects of data model and query language, which can be used for future studies of other models and languages.