Most of us have been taught that the database writers wake up every 3 seconds. I decided to see if this was true on an Oracle Database 11g system. Figure 6-23 is a snippet of the results. I started an Oracle Database 11g instance, placed no load on the system, and operating system traced the database writer. Notice the sleep time is just about 3 seconds and the system call is semtimeodop. If you recall from Chapter 3, when a server process sleeps during latch acquisition, it's because it issued a select system call. The select does not allow the process to be woken, but a semaphore call does. This is an important distinction, because the database writer may need to be woken up for a variety of reasons, such as a checkpoint or a free buffer waits event (described a little later in the "Free Buffer Waits" section). If you were to trace an active database writer process, you would see frequently occurring write-related system calls, as shown in Figure 6-24.

Figure 6-23. This is an operating system trace of an Oracle Database 11g database writer on a very idle system. Notice the 3-second database writer sleep is induced by a semaphore call. This allows the database writer to be woken if necessary.

Figure 6-24. This is an operating system trace of an Oracle Database 11g database writer on a very DML-intensive system. No lines where removed between the write system call (pwrite64). Notice the database writer made a variety of different-sized writes.