Most Oracle sites will never experience mutex contention and if they do, the stress will probably be related to pinning a cursor in either shared or exclusive mode.

Interestingly, for mutexes to operate, the operating system must support the compare and swap (CAS) operation. Reduced instruction set computer (RISC) operating systems, such as AIX or PA-RISC, may have chosen to reduce their instruction set by eliminating the CAS operation. In situations like these, Oracle will simulate the CAS operation by using a pool of latches (1,024 by default in Oracle Database 11g Release 1). The latches are named KGX, and their number can be changed by altering the instance parameter _kgx_latches. Obviously, this is not optimal for performance, but one hopes the net result will be beneficial.

There are actually a number of mutex-related wait events, as listed in Table 7-1. While I would like all mutex-related wait events to start with mutex, Oracle has taken a different path. The mutexes associated with the library cache all start with the word cursor. It makes sense, since the library cache is full of cursors, but it makes discovering new mutex usage more difficult for the performance analyst.