Figure 6-1 is just one way to model an Oracle data block. Oracle buffers are indeed cached in Oracle's buffer cache. While I will delve deeper into data blocks later in this chapter, my point here is that a buffer simply represents the block on disk. But if a block is accessed, a difference between the buffered block and the on-disk block can occur.

While there are many buffer states, as indicated by the state column in v$bh, they can be practically summarized into three modes: free, dirty, and pinned. The algorithms I'll present in this chapter are closely associated with a buffer's state.

A buffer is free when it matches the block on disk. I commonly refer to a free buffer as a mirrored buffer, because it mirrors what is on disk. Figure 6-2 shows just how simple it is to determine the number of free buffers in the buffer cache. A free buffer may indeed be empty (for example, after an instance restart), but it will most likely contain real block information, such as rows. A free buffer can be replaced without any corruption because there is a copy on disk. Of course, if a transaction commits, then at a minimum, the buffer change must be recorded in an online redo log.