ANBOB™

For Oracle versions 12.1.0.2 up to 19.5, the wait event ‘log file parallel write’ includes both the submission of the IO request (io_submit call) and the waiting for the submitted IOs to finish (io_getevents call). This means the waiting time of the event in this case is the total latency of the IO request, including operating system processing.

Starting from version 12.2, only if the multi-tenant option is used, this is changed, and the wait event ‘log file parallel write’ only is shown if the submitted IOs are not available after they are submitted, and thus the logwriter process has to wait for them. If the wait is shown, the time in the event is the actual waiting time for the IOs to finish from the perspective of the logwriter process, not the IO latency time.

Of course if you have multiple log groups on disks with different latencies, this wait event waits for all IOs to finish, which means it times the slowest IO of all submitted IOs.

Thanks Frits Hoogland show it in his blog.

打赏

微信扫一扫，打赏作者吧～