Department of

Computer Science

COMP-370
Algorithms
Lecture 6
Trees II

Spring Semester
2022
02/03/2022
Demetris Trihinas
Lecture 6 | Trees II 1 1
[email protected]
 Department of
Computer Science

Lecture Overview
• Balanced Trees

• AVL Trees
• Properties

• Maintaining Balance

• Data on Disk

• M-ary and B-trees

02/03/2022
Demetris Trihinas
Lecture 6 | Trees II 2 2
[email protected]
 Department of
Computer Science

Previous Lecture: Binary Search Trees

• Search running is 𝑂(ℎ).

• But… height is at best log 2 𝑛.

• and… height is at worst 𝑛 − 1.

• This means to perform search on BST, best

case is 𝑂(log 2 𝑛) and worst case 𝑂(𝑛).

• Can we tighten bounds and guarantee search is 𝑂(log 2 𝑛)?

02/03/2022
Demetris Trihinas
Lecture 6 | Trees II 3 3
[email protected]
 Department of
Computer Science

Idea: Reduce height of a Tree

• Fill internal nodes:
Can chain looking trees become perfect trees?

• Increase number of children, why have 2 when you could

have 4, 6, 8, …
• Yes, but NOT by too much because then trees end up with large
child lists (same problem… different place).
02/03/2022
Demetris Trihinas
Lecture 6 | Trees II 4 4
[email protected]
 Department of
Computer Science

Balanced Tree
• A tree where its leafs are ALL at the same height.

Not full but Full but not Full, balanced

balanced balanced and perfect

Remember (Lecture 5): Full is a tree where every internal node

has exactly two non-empty children.
02/03/2022
Demetris Trihinas
Lecture 6 | Trees II 5 5
[email protected]
 Department of
Computer Science

AVL Trees (Adelson‐Velskii and Landis)

• An AVL tree is a BST with a balance condition.

• Balance condition must be easily maintained and must

guarantee tree depth is O(logn).

• An AVL tree is identical to a BST, except that for every

node in the tree, the height of the left and right
subtrees can differ by at most 1.

• A full tree is also a AVL tree.

02/03/2022
Demetris Trihinas
Lecture 6 | Trees II 6 6
[email protected]
 Department of
Computer Science

Exercise
• Which one is an AVL tree an why?

02/03/2022
Demetris Trihinas
Lecture 6 | Trees II 7 7
[email protected]
 Department of
Computer Science

Exercise
• Is the following an AVL tree?

Requiring balance at the root is not enough.

02/03/2022
Demetris Trihinas
Lecture 6 | Trees II 8 8
[email protected]
 Department of
Computer Science

The Height of an AVL Tree

• Suppose 𝑁(ℎ) is the minimum number of nodes of an
AVL tree with height ℎ.

• We know: 𝑁(0) = 1 and 𝑁(1) = 2.

• For ℎ ≥ 2 an AVL tree must:

• Have a root,

• One of the two subtrees must have height ℎ − 1.

• The height of the two subtrees can differ at most by 1.

02/03/2022
Demetris Trihinas
Lecture 6 | Trees II 9 9
[email protected]
 Department of
Computer Science

The Height of an AVL Tree

• Thus:
𝑁(ℎ) = 𝑁(ℎ − 1) + 𝑁(ℎ − 2) + 1

• This recursive relationship closely resembles Fibonacci

numbers, from which the bound claimed above on the
height of an AVL tree follows.

• So… the height of an AVL tree with n nodes is O(logn).

02/03/2022
Demetris Trihinas
Lecture 6 | Trees II 10 10
[email protected]
 Department of
Computer Science

AVL Trees
• All operations are now guaranteed to be of O(logn).

• Yes… but insertion?

Inserting key = 6 would destroy
the balance condition at node
with key = 8.

-> left subtree has height 2 while

right is 0.

To preserve balance condition we

6 will be performing rotation.
02/03/2022
Demetris Trihinas
Lecture 6 | Trees II 11 11
[email protected]
 Department of
Computer Science

AVL Tree Implementation

• The coding of an AVL tree is similar to a BST (because AVLs
are also BSTs). But… we add 1 more field:
• The height of the tree from each node.

• The AVLNode:
• key
• height
• Left subtree pointer
• Right subtree pointer

02/03/2022
Demetris Trihinas
Lecture 6 | Trees II 12 12
[email protected]
 Department of
Computer Science

Node Insertion
• The insertion of a node happens as in BSTs but
we keep trace of the path followed.

• Afterwards we must follow the reverse path to

update the height of the nodes with possible
new values.
• If AVL tree has become imbalanced, meaning there is a
node with subtrees that have a height difference >=1 then
we must rebalance the tree The process is called rotation.
02/03/2022
Demetris Trihinas
Lecture 6 | Trees II 13 13
[email protected]
 Department of
Computer Science

Node Insertion
• Insert the values 72, 26, 9 to an empty AVL tree:

To avoid violation,
ideally, we would like
to have this.

02/03/2022
Demetris Trihinas
Lecture 6 | Trees II 14 14
[email protected]
 Department of
Computer Science

Left Rotation
• Insertion into the left subtree of the left child.

• Before insertion: Subtrees R, S and T have same height h.

• After insertion: Suppose node is inserted to subtree R, meaning

its height becomes h+1.
Node added Left Rotation
and heights
updated

02/03/2022
Demetris Trihinas
Lecture 6 | Trees II 15 15
[email protected]
 Department of
Computer Science

Left Rotation Process

• Left rotation of A and B trees means:
1. A.left = B.right
2. B.right = A
3. A.height = C.height + 1
4. B.height = C.height + 2

• Before the rotation, A was parent to B, afterwards, B becomes

parent to A.
• After left rotation, the tree remains an AVL tree with:
A.height = h + 1 (height of R)

02/03/2022
Demetris Trihinas
Lecture 6 | Trees II 16 16
[email protected]
 Department of
Computer Science

Exercise
• Insert 6 to the following AVL tree and make sure
balance condition is preserved:

02/03/2022
Demetris Trihinas
Lecture 6 | Trees II 17 17
[email protected]
 Department of
Computer Science

Exercise
• The following tree is imbalanced. Find where the misbalancing
occurs and fix the problem.
A
Imbalanced
B ℎ𝑙𝑒𝑓𝑡 = 3
C
ℎ𝑟𝑖𝑔ℎ𝑡 = 1

02/03/2022
Demetris Trihinas
Lecture 6 | Trees II 18 18
[email protected]
 Department of
Computer Science

Right Rotation (Symmetric to LR)

• Insertion into the right subtree of the right child.

• Before insertion: Subtrees R, S and T have same height h.

• After insertion: Suppose node is inserted to subtree T, meaning

its height becomes h+1.
Right Rotation

02/03/2022
Demetris Trihinas
Lecture 6 | Trees II 19 19
[email protected]
 Department of
Computer Science

Right Rotation Process

• Right rotation of A and B trees means:
1. A.right = C.left
2. C.left = A
3. A.height = B.height + 1
4. C.height = B.height + 2

• Before the rotation, A was parent to C, afterwards, C becomes

parent to A.
• After right rotation, the tree remains an AVL tree with:
C.height = h + 1 (height of T)

02/03/2022
Demetris Trihinas
Lecture 6 | Trees II 20 20
[email protected]
 Department of
Computer Science

Exercise
• Starting from an initial empty AVL tree, insert the items
3, 2, 1, 4, 5, 6, 7:

First issues comes

when we insert 3

02/03/2022
Demetris Trihinas
Lecture 6 | Trees II 21 21
[email protected]
 Department of
Computer Science

Exercise
Next issue comes when we insert 5 with node 3 right height being 2 and left 0.

02/03/2022
Demetris Trihinas
Lecture 6 | Trees II 22 22
[email protected]
 Department of
Computer Science

Exercise
Next issue comes at the root when we insert 6.

02/03/2022
Demetris Trihinas
Lecture 6 | Trees II 23 23
[email protected]
 Department of
Computer Science

Exercise
Finally the insertion of 7 causes another issue:

02/03/2022
Demetris Trihinas
Lecture 6 | Trees II 24 24
[email protected]
 Department of
Computer Science

Double Rotations
• An insertion to right subtree of left child and an insertion
to left subtree of the right child require two rotations.

• These are the two cases where the insertion occurs on the
“inside” (i.e., left–right or right–left).
Still… imbalanced

Adding to Y cannot be
fixed with single rotation.
02/03/2022
Demetris Trihinas
Lecture 6 | Trees II 25 25
[email protected]
 Department of
Computer Science

Left-Right Double Rotation

• Insertion into the right subtree of the left child.

• Any insertion to the right subtree of 𝑘1 , meaning to

either B or C, will cause imbalance, which requires
double rotation to fix.

02/03/2022
Demetris Trihinas
Lecture 6 | Trees II 26 26
[email protected]
 Department of
Computer Science

Left-Right Double Rotation Process

• Left-Right rotation involving 𝑘1 , 𝑘2 , 𝑘3 means:
1. k1.right = k2.left

2. k3.left = k2.right

3. k2.left = k1

4. k2.right = k3

02/03/2022
Demetris Trihinas
Lecture 6 | Trees II 27 27
[email protected]
 Department of
Computer Science

Right-Left Double Rotation

• Insertion into the left subtree of the right child.

• Any insertion to the left subtree of 𝑘2 , meaning to

either B or C, will cause imbalance, which requires
double rotation to fix.

02/03/2022
Demetris Trihinas
Lecture 6 | Trees II 28 28
[email protected]
 Department of
Computer Science

Right-Left Double Rotation Process

• Right-Left double rotation involving 𝑘1 , 𝑘2 , 𝑘3 means:
1. k3.left = k2.right

2. k1.right = k2.left

3. k2.left = k1

4. k2.right = k3

02/03/2022
Demetris Trihinas
Lecture 6 | Trees II 29 29
[email protected]
 Department of
Computer Science

Exercise
• Insert in the following AVL tree the values 16, 15, 14,
13, 12, 11, 10 , 8 and 9:

02/03/2022
Demetris Trihinas
Lecture 6 | Trees II 30 30
[email protected]
 Department of
Computer Science

Exercise
• Inserting 15 causes a height imbalance at node 7:

Adding to left of the

right subtree.

02/03/2022
Demetris Trihinas
Lecture 6 | Trees II 31 31
[email protected]
 Department of
Computer Science

Exercise
• Inserting 14 causes a double rotation:

02/03/2022
Demetris Trihinas
Lecture 6 | Trees II 32 32
[email protected]
 Department of
Computer Science

Exercise
• Inserting 13 causes a imbalance at the root but single
rotation is enough:

02/03/2022
Demetris Trihinas
Lecture 6 | Trees II 33 33
[email protected]
 Department of
Computer Science

Exercise
• Inserting 12 causes imbalance but single rotation is
enough:

02/03/2022
Demetris Trihinas
Lecture 6 | Trees II 34 34
[email protected]
 Department of
Computer Science

Exercise
• To insert 11, a single rotation needs to be performed, and the
same is true for insertion of 10. We insert 8 without a rotation
creating an almost perfectly balanced tree:

02/03/2022
Demetris Trihinas
Lecture 6 | Trees II 35 35
[email protected]
 Department of
Computer Science

Data on Disk
• Up to now, we have assumed entire data (structure) can fit
in memory -> unrealistic.

• In reality we must store data on disk and only parts of the

dataset resides in memory.

• But… disk I/O magnitudes of scale slower than processor

(e.g., 120 disk I/O equal billions of ops by processor).
• So… Big-Oh analysis fails as not all operations are equal.

02/03/2022
Demetris Trihinas
Lecture 6 | Trees II 36 36
[email protected]
 Department of
Computer Science

Data on Disk
• Since disk I/Os are slow… we must reduce disk I/Os.

• Example… Florida citizen driving records – 10M records

• Driving records (256 bytes) with driving ID the key (32 bytes).
• 10M x 256 = 2.5GB (assume… data cannot fit in memory)
• 20 employees accessing system… meaning each employee
has roughly access to 1/20th of server resources.
• So… in 1 second we can execute billions of ops or perform 6
disk I/Os.

02/03/2022
Demetris Trihinas
Lecture 6 | Trees II 37 37
[email protected]
 Department of
Computer Science

Trees for Indexes

• If we consider BST… it may be unbalanced meaning worst
case linear search -> 10M disk accesses needed.

• If we consider AVL tree which guarantees height O(logN)

then logN = 25 disk access and thus 4s.
• So… reducing the tree height helps!
• Also.. If we have more branching, we have less height.
• Perfect binary tree with 31 nodes has height of 5 while a 5-ary
tree has height of 3… remember M-ary tree height 𝑂(log 𝑀 𝑁).

02/03/2022
Demetris Trihinas
Lecture 6 | Trees II 38 38
[email protected]
 Department of
Computer Science

M-ary Trees
• For a binary tree we need 1 key to decide which of two
branches to follow.

• For a M-ary tree we need M-1 key to decide.

5-ary tree with 31 nodes

02/03/2022
Demetris Trihinas
Lecture 6 | Trees II 39 39
[email protected]
 Department of
Computer Science

B-trees
• A B-tree of order M is an M-ary tree with these properties:
• The data items are stored at leaves.
• The non-leaf nodes store up to 𝑀 − 1 keys to guide the searching
and key 𝑖 represents the smallest key in subtree 𝑖 + 1.
• The root is either a leaf or has between two and M children.
• All non-leaf nodes (except the root) have between ⌈𝑀/2⌉ and 𝑀
children.
• All leaves are at the same depth and have between ⌈𝐿/2⌉ and 𝐿
data items.

02/03/2022
Demetris Trihinas
Lecture 6 | Trees II 40 40
[email protected]
 Department of
Computer Science

B-tree of order 5

• Non-leaf nodes have between 3 and 5 children and thus…

between 2 and 4 keys.
• The root has 3 children but it can possibly have two children.
• Here… L = 5 meaning each leaf has between 3 to 5 data items.
02/03/2022
Demetris Trihinas
Lecture 6 | Trees II 41 41
[email protected]
 Department of
Computer Science

B-tree
• Usually, each node represents one disk block that needs 1
disk I/O to be accessed.

• M and L are chosen based on items that can fit disk block.

• Suppose block is 8192 bytes and key (e.g., driver ID) is 32

bytes. For a M-ary B-tree we have M-1 keys, for a total of
(M-1)*32 bytes plus M*4 bytes for branch pointers.

• So the largest value of M for which 36*M - 32 fits in 8192 is

M = 228. Also, L = 8192/256 = 32 data records.
02/03/2022
Demetris Trihinas
Lecture 6 | Trees II 42 42
[email protected]
 Department of
Computer Science

B-tree Insertion
• Insert 57 to B-tree:

57

02/03/2022
Demetris Trihinas
Lecture 6 | Trees II 43 43
[email protected]
 Department of
Computer Science

B-tree Insertion
• Insert 55 to B-tree

But… leaf is already full.

02/03/2022
Demetris Trihinas
Lecture 6 | Trees II 44 44
[email protected]
 Department of
Computer Science

B-tree Leaf Splitting

• When a leaf is at L+1 then it is split into 2 leafs both
guaranteed to have the min number of data records.

• The parent must also be updated and the mid of the

previous leaf is used as key.

• So… two disk accesses are required to write these leafs

and 1 more for updating parent.

02/03/2022
Demetris Trihinas
Lecture 6 | Trees II 45 45
[email protected]
 Department of
Computer Science

B-tree Insertion
• Insert 55 to B-tree:

02/03/2022
Demetris Trihinas
Lecture 6 | Trees II 46 46
[email protected]
 Department of
Computer Science

B-tree Insertion
• Insert 40 to B-tree:

But… non-leaf is also

already full.

Leaf is already full.

02/03/2022
Demetris Trihinas
Lecture 6 | Trees II 47 47
[email protected]
 Department of
Computer Science

B-tree Non-Leaf Splitting

• When a parent is split, we must update the values of the
keys and also the parent’s parent.
• So… this costs 5 disk I/Os.

• When a non-leaf node is split, its parent always gains a

child.
• What if the parent already has reached its limit of children?
• In that case, we continue splitting nodes up the tree until either
we find a parent that does not need to be split or we reach the
root.
02/03/2022
Demetris Trihinas
Lecture 6 | Trees II 48 48
[email protected]
 Department of
Computer Science

B-tree Insertion
• Insert 40 to B-tree:

02/03/2022
Demetris Trihinas
Lecture 6 | Trees II 49 49
[email protected]
 Department of
Computer Science

COMP-370
Algorithms

Questions?

02/03/2022
Demetris Trihinas
Lecture 6 | Trees II 50 50
[email protected]