Data Processing

Abstract

AnalyticsDB processes queries with the help of eight different database operators. Each database operator O accepts as input a single or several column(s) C and/or (a) position list(s) P. Each operator O evaluates a condition D and outputs a single value or column(s) or a position list. A position list can be seen as a filter on a column, as it references a subset of the total entries in a column. Since AnalyticsDB applies the pattern of late materialization (see Sect. 3.2), it tries to work with position lists as long as possible during the processing of a query for performance improvements. Table 5.1 presents an overview of the operators. Although the table shows that most operators accept one or more columns as input, they also accept a position list as input if the to be processed column is not materialized yet. Similar to the input, the output can also be a materialized column or just a position list. A database table from an operator’s perspective is just a collection of columns.