Review

Aggregation is an important, and widely used, feature of SQL, especially in decision support queries. Over the years, however, much doubt has been expressed about the ad hoc nature of grouping/aggregation in SQL, and the fact that certain conceptually simple queries can be represented only in convoluted ways in SQL. By and large, proposed improvements have also been ad hoc, without a careful understanding of the class of queries better expressed in the "improved" languages.

The interesting observation made by Chatziantoniou and Ross in this paper is that a large class of conceptually simple, and efficiently evaluable, queries involves correlated aggregations over the same groups. This observation led them to formulate an SQL extension with the following key insights:

• a sequence of grouping variables, and predicates correlating these variables, can be used to naturally represent a nested sequence of aggregations and selections over the same groups (for example, identifying, for each customer, their longest call and the corresponding area code), and
• performing aggregations and selections repeatedly over the same groups can be done much more efficiently than using standard joins and aggregations.

This paper deserves to be widely read for the simplicity and elegance of its solution for enhancing support for grouping/aggregation in SQL.

[Full Disclosure: The authors of this paper are former co-authors. However, this has not affected my views on the paper.]