How To Build A Performance Indicator Using Machine Learning (Connect 2019)

Both a time based (think AWR) and a sample based (think ASH) performance analysis provide the Oracle performance specialist with an incredibly rich data set and structured analysis possibilities. A DBA/Developer proficient in both AWR and ASH analysis is highly valuable in today's market. However, complex performance situations are not always quickly diagnosable using an AWR data time based analysis or by an ASH incident analysis. It's because situations are sometimes a complex mix of highly fragmented time classifications and a changing workload mix. This makes diagnosis more time consuming and complex. And certainly, the diagnosis will not be completed the second the SLA breach occurs! A solution is to use an unsupervised classification machine learning (UML) model to immediately tag, for example, a snapshot duration as bad, ok, or wonderful. A UML predictive model has the ability to combine a virtually unlimited number of performance related characterists into a small set of classes, such as green, yellow, or red. In summary, this is done by tagging a few of the known problem time samples as "bad", training the UML model, then noting the assigned classification for the "bad" samples. Then we integrate a new sample into the model and observe how the model classifies it. If the new sample class is "bad" then we have identified a potential performance situation. Now, any number of actions can occur, such as an email, text, flashing red lights, sirens, etc. This "new sample integration and alert" process can occur in less than a second, when automated. It happens well before the phone rings or a ticket is submitted, enhancing IT business operational efficiency. Join me as I demonstrate how a performance focused Oracle DBA/Developer can use machine learning to their advantage!