querying v$system_event, even at 10 second intervals should be cheap. You say it "is very CPU consuming". Have you quantified the amount of CPU it uses? S-ASH, or simulated ASH, used about 1% of 1 CPU when I had 10 session active an that was polling every second. In S-ASH default install I poll every 3 seconds and if you are using version 10g, I just query v$session instead of joining to v$session_wait so it is even cheaper. If you need something cheaper, for some reason, then reading the SGA is the way to go, but in your example, I can't imagine any reason to sample v$system_event that fast. On the other hand, sampling v$session on 10g to reproduce ASH is very apealing - it avoids the license issue of ASH and it provides all the rich performance data, multi dimensional aggregations and drilldowns we get from ASH. (v$system_event is pre-aggregated data that provides no possibilites of drilling into the sesions/users involved, there sql nor the detailed p1,p2,p3 needed to solve the majority of wait event problems. Best Wishes Kyle Hailey http://perfvision.com -- //www.freelists.org/webpage/oracle-l