Oracle9i Release 2 and later have multiple redo allocation latches available, and the redo copy latches are normally not required. In summary, the redo log buffer works like this: A server process contends for one of the redo allocation latches, acquires one, allocates redo log buffer space, copies its redo into the redo log buffer, and releases its redo allocation latch. In more detail, it works like this:

* The server process gets the memory address of the last allocation pointer, which is marker MK100 in Figure 8-3.

* The server process adds the redo bytes it needs to marker MK100 and moves the pointer clockwise, perhaps to marker MK200.