Non-linear Optimization of Performance Functions for Autonomic Database Performance Tuning

Modern ondemand environments are coined by a heterogeneous diversity of components, architectures and applications. High performance, availability and further service level agreements need to be satisfied under any circumstances in order to please customers. Today, highly skilled database administrators (DBAs) are required to tune the DBMS within their complex environments. Achieved DBMS' performance depends on individual DBA skills, home-grown tuning scripts and in most cases is reactive to obvious and urgent performance problems. This paper addresses the idea of classifying, formalizing, obtaining, storing, maintaining, exchanging and individually adapting DBA expert tuning-knowledge as shared domain of understanding in the autonomic management process. Hereby, we focus our attention on the development of a resource dependency model that allows for (precise) optimization and decision-support at run-time, in contrast to traditional trial-and- error, feedback-based tuning methodologies based on best- practices.