Partitioning Algorithms for the Computation of Average Iceberg Queries

Iceberg queries are to compute aggregate functions over an attribute (or set of attributes) to nd aggregate values above some speci ed threshold. It's di cult to execute these queries because the number of unique data is greater than the number of counter buckets in memory. However, previous research has the limitation that average functions were out of consideration among aggregate functions. So, in order to compute average iceberg queries e ciently we introduce the theorem to select candidates by means of partitioning, and propose POP algorithm based on it. The characteristics of this algorithm are to partition a relation logically and to postpone partitioning to use memory e ciently until all buckets are occupied with candidates. Experiments show that proposed algorithm is a ected by memory size, data order, and the distribution of data set.