As with most latches, it is very common for the operating system to be experiencing a raging CPU bottleneck. While increasing CPU capacity is a valid solution and will most likely enable more work to flow through Oracle, simply adding more redo allocation latches to the Oracle system will most likely solve the problem, without requiring any operating system or application-related changes.

If a process must acquire one of the redo copy latches, it needs to have this control structure to copy redo into the redo log buffer. This type of latch contention is unusual, but can occur with very high-concurrency OLTP workloads.

For pre-Oracle Database 10g systems, the wait event will be latch free, with the associated latch being redo copy. For later systems, the wait event is simply latch free: redo copy. Regardless of your Oracle release, a simple and very effective solution is to increase the instance parameter _log_simultaneous_copies, which will increase the number of redo copy latches. This is an elegant solution that is very likely to eliminate redo copy latch contention without requiring any changes to either the operating system or to the application.