Skip List From Scratch: A Guide

The Preface and Base Structure of a Skip List

A skip list is a probabilistic data structure that serves as a dynamic set, offering an alternative to red-black or AVL trees. It supports operations like search, inserts, and delete, which typically have an average time complexity of O(log(n)), although in the worst cases, time complexity will be O(n). Skip lists are memory buffers in various NoSQL solutions, including Redis and RocksDB.

The structure of a skip list is composed of multiple layers, referred to as levels. Each level contains roughly half the nodes of the previous level. The first level includes all sorted nodes, and each subsequent level contains about half of the nodes from the level below. This hierarchical setup helps visualize the basic structure of a skip list.

As you can see, our skip list consists of the levels. All the values of each skip list are sorted. Also, we must think about the high bound of levels. We take the const 16 because the original paper refers to it, and it ensures log(n) time complexity for lists that contain 2^16 elements. Let’s code the base structure of our skip list.

const (
	maxLevel    = 16
	probability = 0.5
)

type node[T any] struct {
	value T
	next  []*node[T]
}

type SkipList[T any] struct {
	compare func(a, b T) int // returns -1 if a is less than b, 0 if a and b are equal, and 1 if a is greater than b
	len     int // number of elements of a skip list
	level int // current level
	head  *node[T] // dummy node
}

func newNode[T any](value T) *node[T] {
	return &node[T]{
		value: value,
		next:  make([]*node[T], maxLevel),
	}
}

func New[T any](compare func(a, b T) int) *SkipList[T] {
	return &SkipList[T]{
		compare: compare,
		level:      1,
		head:       newNode[T](*new(T)),
	}
}

This code requires an explanation. Our node has a field next that refers to the node on all the levels. Also, we can denote the function compare that returns -1 if a is less than b, 0 if a is equal to b, and 1 if a is greater than b. We initialize our skip list by 1 level.

Search Operation

The search operation of the skip list is very simple. We will follow by the highest level before our node is less than the current node; then we down to the next level and so on. We don’t go to the first level. This image illustrates our route. We want to find the node with the value 12. We have to pass through two nodes at level 3, one node at level 2, and two nodes at level 1.

In our implementation, we will start from the highest level of our skip list and will go unless the current node is smaller than the desired node. We will repeat our doing until we get to level 1.

func (s *SkipList[T]) Search(value T) (T, bool) {
	current := s.head

	for i := s.level - 1; i >= 0; i-- {
		for current.next[i] != nil && s.compare(current.next[i].value, value) < 0 {
			current = current.next[i]
		}
	}

	if current.next[0] != nil && s.compare(key, current.next[0].value) == 0 {
		return current.next[0].value, true
	}

	return *new(T), false
}

Insert Operation

The insertion operation is somewhat more complex than the search operation. We introduce a new update variable that records the last node smaller than the inserted value at each level. If the inserted node is already present in the skip list, no action is taken. Next, we need to determine the level to which the new node will be promoted. If this calculated level exceeds the current level, additional levels will be added. We then update the references.

The image below illustrates the insertion operation. The node with the value 1 is a dummy node. The current level is 3, but after calling randomLevel, a new level of 4 is determined. The highlighted node is the newly inserted node.

The code of the insert operation.

func (s *SkipList[T]) Insert(value T) {
	current := s.head
	update := make([]*node[T], maxLevel)

	for i := s.level - 1; i >= 0; i-- {
		for current.next[i] != nil && s.compare(current.next[i].value, value) < 0 {
			current = current.next[i]
		}
		update[i] = current
	}

	if current.next[0] != nil && s.compare(current.next[0].value, value) == 0 {
		return
	}

	level := randomLevel()

	if level > s.level {
		for i := s.level; i < level; i++ {
			update[i] = s.head
		}
		s.level = level
	}

	x := newNode[T](value)

	for i := 0; i < level; i++ {
		x.next[i] = update[i].next[i]
		update[i].next[i] = x
	}

	s.len++
}

func randomLevel() int {
	level := 1
	for rand.Float64() < probability && level < maxLevel {
		level++
	}
	return level
}

Delete Operation

The delete operation is similar to the insert operation. We use the update variable to store the last node smaller than the node to be deleted. After identifying all relevant nodes, we remove them if the subsequent node is the one to be deleted. If the dummy node points to nil, we will decrement the level.

func (s *SkipList[T]) Remove(value T) bool {
	current := s.head
	update := make([]*node[T], maxLevel)

	for i := s.level - 1; i >= 0; i-- {
		for current.next[i] != nil && s.compare(current.next[i].value, value) < 0 {
			current = current.next[i]
		}
		update[i] = current
	}

	if current.next[0] == nil || s.compare(current.next[0].value, value) != 0 {
		return false
	}

	for i := 0; i < s.level; i++ {
		if update[i].next[i] == nil || s.compare(current.next[i].value, value) != 0 {
			break
		}
		update[i].next[i] = update[i].next[i].next[i]
	}

	for s.level > 1 && s.head.next[s.level-1] == nil {
		s.level--
	}

	s.len--

	return true
}

In this article, we won’t consider such operations as ceil, floor, etc.

As you can see, the implementation is easier than that of balanced trees. However, it requires more memory and cannot guarantee logarithmic time complexity.