Insertions and deletions

in B+ trees
Introduction to Database Design 2011, Lecture 11
Supplement to lecture slides

Rasmus Ejlers Møgelberg

 Initial setup
• We consider the B+ tree below

judy rick

bob jane mike pete tom

abe al bob eddie jane joe judy karen mike nan pete phil rick sol tom

Rasmus Ejlers Møgelberg

2
 Observations
• Observe that the tree has fan out 3

• Invariants to be preserved
- Leafs must contain between 1 and 2 values
- Internal nodes must contain between 2 and 3 pointers
- Root must have between 2 and 3 pointers
- Tree must be balanced, i.e., all paths from root to a leaf
must be of same length

Rasmus Ejlers Møgelberg

3
 Inserting Vince

judy rick

bob jane mike pete tom

abe al bob eddie jane joe judy karen mike nan pete phil rick sol tom vince

Rasmus Ejlers Møgelberg

4
 Inserting Vera
• Leaf consisting of tom and vince is split and extra
pointer is inserted in parent

judy rick

bob jane mike pete tom vince

abe al bob eddie jane joe judy karen mike nan pete phil rick sol tom vera vince

Rasmus Ejlers Møgelberg

5
 Inserting rob
• Inserting rob is more difficult. We first create a
new leaf node and insert it as below

• The node above is temporarily extended to

contain 4 pointers
judy rick

bob jane mike pete sol tom vince

abe al bob eddie jane joe judy karen mike nan pete phil rick rob sol tom vera vince

Rasmus Ejlers Møgelberg

6
 Inserting rob
• The overfull internal node is then split in 2

• The new pointer is inserted into the root node

which then becomes overfull

judy rick tom

bob jane mike pete sol vince

abe al bob eddie jane joe judy karen mike nan pete phil rick rob sol tom vera vince

Rasmus Ejlers Møgelberg

7
 Inserting rob
• Finally the overfull root is split in 2

• At this point the tree satisfies the requirements of

slide 3 and so the insertion procedure ends
rick

judy tom

bob jane mike pete sol vince

abe al bob eddie jane joe judy karen mike nan pete phil rick rob sol tom vera vince

Rasmus Ejlers Møgelberg

8
 Deleting jane
• Is straight forward

• Note that the node above the leaf where jane was
deleted must also be updated
rick

judy tom

bob joe mike pete sol vince

abe al bob eddie joe judy karen mike nan pete phil rick rob sol tom vera vince

Rasmus Ejlers Møgelberg

9
 Deleting bob

rick

judy tom

eddie joe mike pete sol vince

abe al eddie joe judy karen mike nan pete phil rick rob sol tom vera vince

Rasmus Ejlers Møgelberg

10
 Deleting joe
• Leads to a leaf being deleted and the parent being
updated

rick

judy tom

eddie mike pete sol vince

abe al eddie judy karen mike nan pete phil rick rob sol tom vera vince

Rasmus Ejlers Møgelberg

11
 Deleting eddie
• Leads to deletion of a leaf

• At this point the parent becomes underfull

rick

judy tom

mike pete sol vince

abe al judy karen mike nan pete phil rick rob sol tom vera vince

Rasmus Ejlers Møgelberg

12
 Deleting eddie
• When a node becomes underful the algorithm will
try to redistribute some pointers from a
neighbouring sibling to it.

• Since this is possible in this case we do it

rick

mike tom

judy pete sol vince

abe al judy karen mike nan pete phil rick rob sol tom vera vince

Rasmus Ejlers Møgelberg

13
 Deleting abe and al
• Leads to deletion of a leaf

• This makes the parent underfull.

• We cannot redistribute pointers again since this

will make the neighbour underfull
rick

mike tom

pete sol vince

judy karen mike nan pete phil rick rob sol tom vera vince

Rasmus Ejlers Møgelberg

14
 Deleting abe and al
• Instead we must merge with the neighbouring
sibling

• But this makes the parent underfull

rick

tom

mike pete sol vince

judy karen mike nan pete phil rick rob sol tom vera vince

Rasmus Ejlers Møgelberg

15
 Deleting abe and al
• Since we can not solve this problem by
redistributing pointers we must merge siblings
again

rick tom

mike pete sol vince

judy karen mike nan pete phil rick rob sol tom vera vince

Rasmus Ejlers Møgelberg

16
 Deleting abe and al
• Since the root is underfull it can be deleted

• The resulting tree satisfies the requirements and

so the deletion algorithm ends

rick tom

mike pete sol vince

judy karen mike nan pete phil rick rob sol tom vera vince

Rasmus Ejlers Møgelberg

17
 General remarks
• When a node becomes underfull the algorithm will
try to redistribute pointers from the neighbouring
sibling either on the left or the right

• If this is not possible, it should merge with one of

them

• The value held in an internal node or the root

should always be the smallest value appearing in a
leaf of the subtree pointed to by the pointer after
the value

Rasmus Ejlers Møgelberg

18