Figure 6-3. The number of dirty buffers can be determined by issuing a simple v$bh-based query looking at the dirty column.

When a buffer is pinned, it cannot be replaced. Another way to look at pinning is as kind of an unofficial lock on a buffer. Since a buffer is not a relational structure, standard locking mechanisms do not apply. Pinning relates to a specific buffer, whereas latches or mutexes can control access to an entire group of buffers. Pinning can be used in conjunction with latching and locking to ensure the proper serialization, protection, and concurrency control are achieved.

Suppose you are a server process that is reading a row in a buffer. It would be extremely rude of someone to replace that specific buffer with another buffer while you were still accessing the row. It would be like you were reading a book, and someone saying, "Hey, let me see that!" and ripping it out of your hands. In fact, placing your hand on a book is a great picture of pinning a buffer. Whenever there is at least one hand on the book, then no other process can take the book away. Many processes can pin the same buffer (read the same book), but as long as one process has the buffer (the book) pinned, it cannot be replaced. When a free buffer's row is being queried, its status changes from free, to pinned, to free again. When a row in a free buffer is changed, its buffer status changes from free, to pinned, to dirty.