Cairo University

Faculty of Computers and Artificial Intelligence

Computer Science Department

Advanced Data Structures B-Trees Dr. Amin Allam

[For more details, refer to “Data Structures and Algorithms in C++” by Adam Drozdek]

1 2-3 trees
Instead of trying to balance a binary search tree by incorporating the red-black properties, we can
relax the binary requirement by allowing some nodes to have 3 children, which allows to keep all
leaves in the same level, which is a more straightforward property that leads to balancing the tree.

A 2-3 tree has the following properties:

• Each internal node has 2 or 3 children.
• Each node has 1 or 2 keys.
• All leaves are on the same level.
• The number of keys in each internal node = the number of its children −1.
• The keys in each node are in ascending order. The keys in the first i children of an internal node
are smaller than its ith key, while the keys in remaining children are larger.

- To insert a new key: The new key is added to the suitable leaf in the correct place to keep all keys
in this leaf sorted. Now if the leaf contains 3 keys, it is split into two leaves: one contains the
smallest key, the other contains the largest key. The middle key is moved to parent. Now if parent
contains 3 keys, this step is repeated. If there is no parent, a new root is created.

- To delete a key from a leaf : After deleting the key, if the leaf is empty and it has a sibling leaf
having 2 key, one of these 2 keys is moved into parent, and the in-between key from parent is
moved to the empty leaf . If it has a sibling leaf having 1 key, they are merged along with the
in-between key from parent into one leaf . If parent becomes empty, this step is repeated unless it
is the root it is just deleted.

- To delete a key from an internal node: the key to be deleted is replaced by its immediate prede-
cessor or successor, which can only exist in a leaf , then the previous procedure is performed.

Since the number of levels changes only by creating a new root or deleting the existing root, all
leaves always remain at the same level. Also, all other 2-3 tree properties are maintained.

To analyse the complexity of searching a 2-3 tree, we look into the maximum possible number
of levels h of a 2-3 tree in terms of the number of existing keys n. The maximum h is achieved
when the number of keys per node is minimum (which is 1 key per node). The number of keys in
the first level (root) is ≥ 1. Since the root must have at least 2 children, the number of keys in
the second level is ≥ 2. Since each of these 2 nodes must have at least 2 children, the number
of keys in the third level is ≥ 4. Similar reasoning leads to conclude that the number of keys
n ≥ 20 + 21 + 22 + · · · + 2h−1 = 2h − 1. Hence, h ≤ log2 (n + 1).

1
FCAI-CU AdvDS B-Trees Amin Allam

2 2-3 tree example:

Starting from an initial empty 2-3 tree, draw all intermediate trees and draw the tree after each of
the following operations:
Insert(M), Insert(T), Insert(F), Insert(Q), Insert(P), Delete(F), Delete(Q), Delete(T).

2.1 Insert(M), Insert(T), Insert(F) M

M M T F M T
F T

2.2 Insert(Q), Insert(P)

M M M Q

F Q T F P Q T F P T

2.3 Delete(F)

M Q Q Q

P T M P T M P T

2.4 Delete(Q)

T P

M P M P T M T

2
FCAI-CU AdvDS B-Trees Amin Allam

2.5 Delete(T)

P
M P
M M P

3 B-trees
B-trees can be viewed as a generalization of 2-3 trees. That is, a 2-3 tree is a B-tree of order 3.
A B-tree of order m has the following properties:
• The root has between 1 and m−1 keys. m
• Each internal node (except root) has between
m 2
and m children.
• Each node (except root) has between 2 −1 and m−1 keys.
• All leaves are on the same level.
• The number of keys in each internal node = the number of its children −1.
• The keys in each node are in ascending order. The keys in the first i children of an internal node
are smaller than its ith key, while the keys in remaining children are larger.

- To insert a new key: The new key is added to the suitable leaf in the correct place to keep all keys
in this leaf sorted. Now if the leaf contains m keys: its middle key is moved to the parent, and the
remaining keys are split into two leaves. Now if parent contains m keys, this step is repeated. If
there is no parent, a new root is created.

: After deleting the key, if the leaf contains m2 −2 keys and it has a
 
- To delete a key froma leaf
sibling leaf having ≥ m2 keys, they are merged along with the in-between key from parent, their


middle key is moved to parent,

 m  and the remaining keys are distributed to the two leaves. If it has a
sibling leaf having only 2 −1 keys,  they are merged along with the in-between key from parent
m
into one leaf . Now if parent has 2 −2 keys, this step is repeated unless it is the root.
- To delete a key from an internal node: the key to be deleted is replaced by its immediate prede-
cessor or successor, which can only exist in a leaf , then the previous procedure is performed.

A similar analysis to the 2-3 tree leads to conclude that the number of levels h in B-tree of order m is
approximately logd m e (n) where n is the number of keys. Therefore, h decreases with the increase
2
of m. This makes a B-tree with large enough m to be very suitable to be stored on secondary
devices (hard drives), where the overhead of accessing a new node is much more than the overhead
of accessing other keys in the same node. That is because nodes are generally not contiguous on
the drive, hence a costly seek is required to move from one node to another. In contrast, keys in the
same node are stored contiguously on the drive, hence little overhead is required to access all keys
in the same node. Moreover, B-trees are often more efficient than red-black trees even if they are
stored in main memory since they significantly reduce cache misses, but they require more storage
than red-black trees since most B-tree nodes contain unused spaces.

3
FCAI-CU AdvDS B-Trees Amin Allam

4 B-tree example:
Starting from an initial empty B-tree of order 5, draw all intermediate trees and draw the tree after
each of the following operations:
Insert(G), Insert(I), Insert(B), Insert(J), Insert(C), Insert(A), Insert(K), Insert(E), Insert(D),
Insert(S), Insert(T), Insert(R), Insert(L), Insert(F), Insert(H), Insert(M), Insert(N), Insert(P),
Insert(Q), Delete(E), Delete(F), Delete(G), Delete(K)

4.1 Insert(G), Insert(I), Insert(B), Insert(J) - Insert(C) -

Insert(A), Insert(K), Insert(E) - Insert(D)
G G C G
B G I J

B C I J A B C E I J K A B D E I J K

4.2 Insert(S) - Insert(T)

C G C G K

A B D E I J K S A B D E I J S T

4.3 Insert(R), Insert(L), Insert(F), Insert(H) - Insert(M)

C G K C G K R

A B D E F H I J L R S T A B D E F H I J L M S T

4.4 Insert(N), Insert(P) - Insert(Q)

C G K R C G K N R

A B D E F H I J L M N P S T A B D E F H I J L M P Q S T

C G N R

A B D E F H I J L M P Q S T

4
FCAI-CU AdvDS B-Trees Amin Allam

4.5 Delete(E) - Delete(F)

K K

C G N R C G N R

A B D F H I J L M P Q S T A B D H I J L M P Q S T

C H N R

A B D G I J L M P Q S T

4.6 Delete(G) - Delete(K)

K K

C H N R C N R

A B D I J L M P Q S T A B D H I J L M P Q S T

C K N R C J N R

A B D H I J L M P Q S T A B D H I L M P Q S T

5 B+ -trees
A B+ -tree is a B-tree where all keys exist in leaves. Keys in intermediate nodes are used only as
separators and for directing search queries. Additional links exist from each leaf to its right sibling.
The purpose of such augmentations is to facilitate range queries, especially for secondary storage.
Once a search query reaches a leaf , all subsequent records can be accessed without accessing
nodes at higher levels.
K

C G N R

A B C D F G H I J K L M N P Q R S T

A static version of B+ -tree is called multi-level indexing. If data is static and no updates are
required, data can be initially sorted once to construct the deepest B+ -tree level, then higher levels
are constructed statically. All links are substituted by formulas as functions of a fixed node size.