Dynamic Oracle Performance Analytics Using Normalized Metrics

AWR collects and persists thousands of performance metrics every hour; the problem is: the root causes of performance anomalies are difficult to detect and it is difficult to know which metrics/attributes are important for DBA’s to focus on to inform their root cause analysis and solutions. A full understanding of these thousands of metrics is impossible, thus many DBA’s (and even the off-the shelf tools) monitor only a standard set of well-known metrics. This “small model” approach may cause you to miss important system behavior or configuration that is relevant to the root cause of the performance problem. The presenters approach to this “small model” meta problem is to massively expand and dynamically extract only the relevant performance metrics. This paradigm shift is achieved by normalizing the data and looking across a wide array of AWR performance metrics gathered during the problem interval. Querying across the normalized data and using statistical analysis, this approach flags unusual trends. By targeting the right metrics at the right time, you “bundle” relevant results which results in event focused and actionable intelligence on the performance issue, and a richer insight into possible solutions.