As with the database writer background process, the log writer background process sleeps by issuing a semaphore call. (As you learned in Chapter 3, when a server process sleeps, in contrast to both the database and log writer background processes, it issues a select call.) This allows another process to wake up the log writer, triggering it to flush the general redo log buffer. For example, if the database writer background process wants to write a dirty buffer to disk, it must ensure any redo associated with the buffer has already been successfully written into an online redo log group. So, the database writer background process will trigger the log writer background process to write and post the log file sync wait event. When the log writer background process is finished writing, the database writer background process can issue its write statement to the operating system.

Figure 8-6. Shown is an operating system trace of an inactive log writer. No commits are being issued. During the snapshot of this Linux Oracle Database 11g Release 1 system, an Oracle session was repeatedly issuing updates, yet even over multiple 3-second intervals, the log writer did not issue a single write request to the operating system.

Diagnosing redo-related performance issues is incredibly straightforward. Oracle's wait interface provides a number of specific redo-related wait events, and the operating system is usually suffering from a severe IO bottleneck. The challenge is usually in being able to make the changes necessary to realize a significant performance gain.