I would make use of Adaptive Thresholds for cases like this. I usually make use of AAS (average active sessions) as a metric Here are good reading materials about the technology http://oracledoug.com/metric_baselines_10g.pdf http://oracledoug.com/adaptive_thresholds_faq.pdf http://karlarao.tiddlyspot.com/#%5B%5BAdaptive%20thresholds%2C%20metric%20baseline%5D%5D Patent here http://www.docstoc.com/docs/56167536/Graphical-Display-And-Correlation-Of-Severity-Scores-Of-System-Metrics---Patent-7246043 This technology has been around since 10g and it makes statistical analysis out of sysmetric/dba_hist_sysmetric views using buckets of time which is called time groups. Of course you need significant data points so it will have better statistical analysis on the workload and if an alert kicks in that means it was able to observe a time period where it is out of the normal workload. This will help you to be proactive on runaway processes or sudden load change. It's pretty helpful on my current environment where we have mixed workloads consolidated on Exadata. And this reminds me of what I'm doing here http://karlarao.tiddlyspot.com/#r2project, I'm actually doing similar stuff where I'm picking buckets of time and making sense of the data points by doing linear regression on AAS with CPU Utilization. Only that Adaptive Threshold is doing it more fine grained by comparing the past buckets (baseline) with the new buckets (new workload)... and it's pretty cool :) -- Karl Arao karlarao.wordpress.com karlarao.tiddlyspot.com -- //www.freelists.org/webpage/oracle-l
