Without sleeping, Oracle processes have no option other than spinning on a latch. And when CPU utilization increases-available CPU cycles are becoming less and less available-we are effectively asking Oracle to consume potentially even more CPU by increased spinning! So, we could be making an already serious situation even worse.

The other problem with increasing the _spin_count value is that all latches will be affected. Key to removing latch contention is focusing on the specific problematic latch. Changing the _spin_count value will affect not only the latch that has problems, but all other latches as well. So, the result is less predictable.

The only time I will increase _spin_count is if the operating system vendor's performance engineer, who must also be an Oracle expert, makes the recommendation and can clearly explain why latch acquisition response time will decrease after changing this value. This is exceptionally rare. If the engineer can't clearly explain why increasing _spin_count will help performance, he is guessing! And guessing is no way to fight performance fires or advance your career.