Mining periodic workload patterns in database audit trails

Information about periodic processing of database operations has a pivotal importance for

continuous physical database design and automated performance tuning of database systems.

This work shows how to detect the oscillations of database workloads caused by the periodical

invocations of user applications. In particular, we present an algorithm for discovering

periodic patterns in the histories of processing of complex and elementary database operations.

In our approach, information collected from the database audit trails is transformed

into a sequence of syntax trees and later on it is compressed in a syntax tree table. The

periodic patterns are discovered through nested iterations over a four dimensional space of

syntax trees and positional parameters of the patterns. Transformations of the patterns are

used to discover the overlaping periodic patterns.