05 — B-Tree vs. B+Tree

The actual difference

	B-Tree	B+Tree
Where values live	in any node, inner nodes included `24:36` `25:01`	only in leaf nodes; inner nodes are pure guide posts `25:01`
Key duplication	each key appears exactly once → more space efficient `25:01`	a key can appear in both a leaf and an inner node
On delete	key disappears from the tree entirely	removed from the leaf, but can be left behind in the inner node `25:31`
Range/sequential scan	must bounce up and down between levels `26:03`	descend once, then rip across the leaf linked list `26:03`
Point query luck	may find the value at the root and stop early `27:30`	always pays the full descent

Sequential access is the game changer. A scan descends once and then never goes back up 26:03 27:58.
Leaving stale keys in inner nodes is free correctness: a deleted key still works as a separator, so why pay to clean it up 25:31 41:38.
Updates are simpler than a red-black tree, which needs rotations 27:30.

"Deletes wouldn't need whole-tree restructuring" — false; B+Trees still restructure on delete 25:45.
"You could keep the non-leaf nodes in memory" — you get that for free from the buffer pool's replacement policy in a B-Tree too, so it is not a differentiator 26:37.