Next: Transactions, Previous: Data, Up: Overview [Contents][Index]
The on disk store is organized in data blocks that correspond the b-tree nodes and the large keys and data.
See The Nodes.
See The Large Stores.
Data blocks are referred by higher level b-tree nodes by their ID (and not by their file offset). A page map associates the storage file offset to block IDs.
In effect, a nonblonde database is a collection of b-trees written one node per block of disk store, with the b-tree nodes referring each other by their block ID and with all block IDs recorded by a page map.
Applications may create and delete data blocks, and may access them by their file offset.
The data blocks affected by changes in the database b-trees are rewritten and relocated.
The data blocks created by applications are not relocated unless modified via the database b-trees or affected by a data store compaction that moves high file offset blocks to lower offsets.