Also notice the statistic value in the second query in Figure 5-6. This is the CPU time consumed (instance statistic CPU used by this session) by the database writer since the instance has started. But it shows zero seconds of CPU consumed! This demonstrates that for the database writer, v$sesstat does not contain background process CPU consumption. Therefore, if you want to classify background process CPU time, you must either gather from the /proc virtual file system or simply query from v$sess_time_model or v$sys_time_model.

Let's shift from background processes to Oracle server processes. Just as with tracing the database writer process, an active server process will make a getrusage call an average of once every 6 seconds. I keep saying active because if a process is waiting for an event to complete (for example, a multiblock read request or an enqueue), the getrusage call obviously cannot be made. When this delay occurs, the time model details are not updated. For example, if a session is waiting 20 seconds for a lock, its time model details will not be updated until after the lock is acquired. Repeated tests demonstrate the delay is sometimes longer, possibly due to the fact that the next getrusage call may not occur immediately after the lock is acquired.

Figure 5-6. This is an example of gathering CPU consumption for the database writer background process. Notice v$sess_time_model contains a value greater than zero, where v$sesstat always will show zero. It appears Oracle is not including the database writer's CPU consumption in the instance CPU consumption statistic. Also, by repeatedly querying v$sess_time_model, the value was observed to be updated about every 6 seconds.