Figure 5-11. Based on the v$event_histogram view and during the report interval, while the average wait time is 14.9 ms (not shown), it can seen that about 25% of the scattered read waits took more than 16 ms. The far-right column is a running total.

Unfortunately, the histogram granularity is limited. The millisecond increment shown in Figure 5-11 is the most detailed time breakdown we can get. This is particularly vexing, because most of the really interesting Oracle wait activity happens around 10 ms. When working in the millisecond neighborhood, a jump from 8 ms to 16 ms is massive, so we lose some valuable diagnostic information. However, the information available is helpful.

One of the more intriguing v$event_histogram applications is gaining an undocumented understanding of Oracle's interworkings. For example, the free buffer wait event (which will be discussed in the next chapter) is documented to cause a 10 ms wait, period. However, if the event histogram view is to be trusted, then why have DBAs seen 15% of the wait times greater than 16 ms? (I have seen this many times.) This indicates there are other things occurring inside Oracle, which we as DBAs are not aware of, forcing us to look a little deeper into Oracle's inner workings.4