• The desired record is found at a shallow depth (few disk accesses). A tree with 256 keys per node can index millions of records in 3 steps or 1 disk access (keeping the root node and next level in memory).

• In general, there are many more searches than insertions.

• Since a node can have a wide range of children, m / 2 to m, an insert or delete will rarely go outside this range. It is rarely necessary to rebalance the tree.

• Inserting or deleting an item within a node is O(blocksize), since on average half the block must be moved, but this is fast compared to a disk access.

• Rebalancing the tree on insertion is easy: if a node would become over-full, break it into two half-nodes, and insert the new key and pointer into its parent. Or, if a leaf node becomes over-full, see if a neighbor node can take some extra children.

• In many cases, rebalancing the tree on deletion can simply be ignored: it only wastes disk space, which is cheap.

Contents    Page-10    Prev    Next    Page+10    Index