“ASH math” of time_waited explained with pictures and simulation
As explained by John Beresniewicz, Graham Wood and Uri Shaft in their excellent overview ASH architecture and advanced usage, avg( v$active_session_history.time_waited ) is not a correct estimate of the average latency (the “true average”) esperienced by a wait event, the reason being that short events are less likely to be sampled. In order to correct this, the authors propose a formula that gives an unbiased estimate of the “true average”.…