One of the limitations of the instance statistics view (for example, v$sysstat) approach is the statistics may be updated only when a call completes. And just as bad, background processes may not have their CPU time recorded (as demonstrated shortly in Figure 5-6). The time system model conquers this problem. To illustrate this, Figure 5-5 is the result of a 5-minute operating system trace of the database writer. Doing some simple math, since the duration is 300 seconds (5 _ 60) and the database writer made 49 getrusage calls, we know that Oracle at least has the capability to update the v$sess_time_model (which feeds into the v$sys_time_model) about once every 6 seconds.

Figure 5-5. Oracle processes call the getrusage system call to determine their resource consumption details. This DBWR operating system trace duration was 300 seconds, which means the database writer background process called getrusage about once every 6 seconds. By the way, notice the DBWR does make some read calls.

Oracle does, in fact, update the v$sess_time_model view with the database writer resource consumption about every 6 seconds. I tested this by simply repeatedly running the first query shown in Figure 5-6. The background cpu time statistic stayed constant for 6 seconds, and then I saw the value incremented. Notice the database writer CPU time is classified as background cpu time, not DB CPU time. Oracle server process time is placed into the DB CPU bucket.