A B+ tree is an m-ary tree with a variable but often large number of children per node. A B+ tree consists of a root, internal nodes and leaves.^[1] The root may be either a leaf or a node with two or more children.

A B+ tree can be viewed as a B-tree in which each node contains only keys (not key–value pairs), and to which an additional level is added at the bottom with linked leaves.

The primary value of a B+ tree is in storing data for efficient retrieval in a block-oriented storage context — in particular, filesystems. This is primarily because unlike binary search trees, B+ trees have very high fanout (number of pointers to child nodes in a node,^[1] typically on the order of 100 or more), which reduces the number of I/O operations required to find an element in the tree.

History

There is no single paper introducing the B+ tree concept. Instead, the notion of maintaining all data in leaf nodes is repeatedly brought up as an interesting variant. Douglas Comer notes in an early survey of B-trees (which also covers B+ trees) that the B+ tree was used in IBM's VSAM data access software, and refers to an IBM published article from 1973.^[2]

Structure

Pointer structure

B+ tree node format where K=4. (p_i represents the pointers, k_i represents the search keys).

As with other trees, B+ trees can be represented as a collection of three types of nodes: root, internal (a.k.a. interior), and leaf. In B+ trees, the following properties are maintained for these nodes:

• If ${\textstyle k_{i}}$ exists in any node in a B+ tree, then ${\displaystyle k_{i-1}}$ exists in that node where ${\displaystyle i\geq 1}$.
• All leaf nodes have the same number of ancestors (i.e., they are all at the same depth).

The pointer properties of nodes are summarized in the tables below:

• K: Maximum number of potential search keys for each node in a B+ tree. (this value is constant over the entire tree).
• ${\displaystyle p_{i}}$: The pointer at the zero-based node index ${\displaystyle i}$.
• ${\displaystyle k_{i}}$: The search key at the zero-based node index ${\displaystyle i}$.

Internal Node Pointer Structure

${\displaystyle p_{0}}$ when ${\displaystyle k_{0}}$ exists	${\displaystyle p_{i}}$ when ${\displaystyle k_{i-1}}$ and ${\displaystyle k_{i}}$ exist	${\displaystyle p_{i}}$ when ${\displaystyle k_{i-1}}$ exists, and ${\displaystyle k_{i}}$ does not exist	${\displaystyle p_{i}}$ when ${\displaystyle k_{i-1}}$ and ${\displaystyle k_{i}}$ do not exist
Points to subtree in which all search keys are less than ${\displaystyle k_{0}}$.	Points to subtree in which all search keys are greater than or equal to ${\displaystyle k_{i-1}}$ and are less than ${\displaystyle k_{i}}$.	Points to subtree in which all search keys are greater than or equal to ${\displaystyle k_{i-1}}$.	Here, ${\displaystyle p_{i}}$ is empty.

Leaf Node Pointer Structure

${\displaystyle p_{i}}$ when ${\displaystyle k_{i}}$ exists	${\displaystyle p_{i}}$ when ${\displaystyle k_{i}}$ does not exist and ${\displaystyle i\neq K}$	${\displaystyle p_{K}}$
Points to a record with a value equal to ${\displaystyle k_{i}}$.	Here, ${\displaystyle p_{i}}$ is empty.	Points to the next leaf in the tree.

Node bounds

The node bounds are summarized in the table below:

Node Type	Min Number of Keys	Max Number of Keys	Min Number of Child Nodes	Max Number of Child Nodes
Root Node (when it is a leaf node)	0	${\displaystyle K}$	0	0
Root Node (when it is an internal node)	1	${\displaystyle K}$	2[1]	${\displaystyle K+1}$
Internal Node	${\displaystyle \lfloor K/2\rfloor }$	${\displaystyle K}$	${\displaystyle \lceil (K+1)/2\rceil }$ ${\displaystyle \equiv \lfloor K/2\rfloor +1}$	${\displaystyle K+1}$
Leaf Node	${\displaystyle \lceil K/2\rceil }$	${\displaystyle K}$	0	0

^[3][4]

Intervals in internal nodes

See also: comparability, total order, and Partially ordered set § Intervals

A simple B+ tree example linking the keys 1–7 to data values d₁-d₇. The linked list (red) allows rapid in-order traversal. This particular tree's branching factor is ${\displaystyle b}$=4. Both keys in leaf and internal nodes are colored gray here.

By definition, each value contained within the B+ tree is a key contained in exactly one leaf node. Each key is required to be directly comparable with every other key, which forms a total order.^[5] This enables each leaf node to keep all of its keys sorted at all times, which then enables each internal node to construct an ordered collection of intervals representing the contiguous extent of values contained in a given leaf. Internal nodes higher in the tree can then construct their own intervals, which recursively aggregate the intervals contained in their own child internal nodes. Eventually, the root of a B+ Tree represents the whole range of values in the tree, where every internal node represents a subinterval.

For this recursive interval information to be retained, internal nodes must additionally contain ${\displaystyle m-1}$ copies of keys ${\displaystyle l_{i}}$ for ${\displaystyle i\in [1,m-1]}$ representing the least element within the interval covered by the child with index i (which may itself be an internal node, or a leaf). Where m represents the actual number of children for a given internal node.

Characteristics

The order or branching factor b of a B+ tree measures the capacity of interior nodes, i.e. their maximum allowed number of direct child nodes. This value is constant over the entire tree. For a b-order B+ tree with h levels of index:

• The maximum number of records stored is ${\displaystyle n_{\max }=b^{h}-b^{h-1}}$
• The minimum number of records stored is ${\displaystyle n_{\min }=2\left\lceil {\tfrac {b}{2}}\right\rceil ^{h-1}-2\left\lceil {\tfrac {b}{2}}\right\rceil ^{h-2}}$
• The minimum number of keys is ${\displaystyle n_{\mathrm {kmin} }=2\left\lceil {\tfrac {b}{2}}\right\rceil ^{h-1}-1}$
• The maximum number of keys is ${\displaystyle n_{\mathrm {kmax} }=b^{h}-1}$
• The space required to store the tree is ${\displaystyle O(n)}$
• Inserting a record requires ${\displaystyle O(\log _{b}n)}$ operations
• Finding a record requires ${\displaystyle O(\log _{b}n)}$ operations
• Removing a (previously located) record requires ${\displaystyle O(\log _{b}n)}$ operations
• Performing a range query with k elements occurring within the range requires ${\displaystyle O(\log _{b}n+k)}$ operations
• The B+ tree structure expands/contracts as the number of records increases/decreases. There are no restrictions on the size of B+ trees. Thus, increasing usability of a database system.
• Any change in structure does not affect performance due to balanced tree properties.^[6]
• The data is stored in the leaf nodes and more branching of internal nodes helps to reduce the tree's height, thus, reduce search time. As a result, it works well in secondary storage devices.^[7]
• Searching becomes extremely simple because all records are stored only in the leaf node and are sorted sequentially in the linked list.
• We can retrieve range retrieval or partial retrieval using B+ tree. This is made easier and faster by traversing the tree structure. This feature makes B+ tree structure applied in many search methods.^[6]

Algorithms

function search(k, root) is
    let leaf = leaf_search(k, root)
    for leaf_key in leaf.keys():
        if k = leaf_key:
             return true
    return false

function leaf_search(k, node) is
    if node is a leaf:
        return node
    let p = node.children()
    let l = node.left_sided_intervals()
    assert ${\displaystyle |p|=|l|+1}$
    let m = p.len()
    for i from 1 to m - 1:
        if ${\displaystyle k\leq l[i]}$:
            return leaf_search(k, p[i])
    return leaf_search(k, p[m])

Deletion

The purpose of the delete algorithm is to remove the desired entry node from the tree structure. We recursively call the delete algorithm on the appropriate node until no node is found. For each function call, we traverse along, using the index to navigate until we find the node, remove it, and then work back up to the root.

At entry L that we wish to remove:

- If L is at least half-full, done

- If L has only d-1 entries, try to re-distribute, borrowing from sibling (adjacent node with same parent as L).

           After the re-distribution of two sibling nodes happens, the parent node must be updated to reflect this change. The index key that points to the second sibling must take the smallest value of that node to be the index key.

- If re-distribute fails, merge L and sibling. After merging, the parent node is updated by deleting the index key that point to the deleted entry. In other words, if merge occurred, must delete entry (pointing to L or sibling) from parent of L.

Note: merge could propagate to root, which means decreasing height.^[9]

B+ tree deletion

Implementation

The leaves (the bottom-most index blocks) of the B+ tree are often linked to one another in a linked list; this makes range queries or an (ordered) iteration through the blocks simpler and more efficient (though the aforementioned upper bound can be achieved even without this addition). This does not substantially increase space consumption or maintenance on the tree. This illustrates one of the significant advantages of a B+tree over a B-tree; in a B-tree, since not all keys are present in the leaves, such an ordered linked list cannot be constructed. A B+tree is thus particularly useful as a database system index, where the data typically resides on disk, as it allows the B+tree to actually provide an efficient structure for housing the data itself (this is described in^[10]: 238 as index structure "Alternative 1").

If a storage system has a block size of B bytes, and the keys to be stored have a size of k, arguably the most efficient B+ tree is one where ${\displaystyle b={\tfrac {B}{k}}-1}$. Although theoretically the one-off is unnecessary, in practice there is often a little extra space taken up by the index blocks (for example, the linked list references in the leaf blocks). Having an index block which is slightly larger than the storage system's actual block represents a significant performance decrease; therefore erring on the side of caution is preferable.

If nodes of the B+ tree are organized as arrays of elements, then it may take a considerable time to insert or delete an element as half of the array will need to be shifted on average. To overcome this problem, elements inside a node can be organized in a binary tree or a B+ tree instead of an array.

B+ trees can also be used for data stored in RAM. In this case a reasonable choice for block size would be the size of processor's cache line.

Space efficiency of B+ trees can be improved by using some compression techniques. One possibility is to use delta encoding to compress keys stored into each block. For internal blocks, space saving can be achieved by either compressing keys or pointers. For string keys, space can be saved by using the following technique: Normally the i-th entry of an internal block contains the first key of block ${\displaystyle i+1}$. Instead of storing the full key, we could store the shortest prefix of the first key of block ${\displaystyle i+1}$ that is strictly greater (in lexicographic order) than last key of block i. There is also a simple way to compress pointers: if we suppose that some consecutive blocks ${\displaystyle i,i+1,...i+k}$ are stored contiguously, then it will suffice to store only a pointer to the first block and the count of consecutive blocks.

All the above compression techniques have some drawbacks. First, a full block must be decompressed to extract a single element. One technique to overcome this problem is to divide each block into sub-blocks and compress them separately. In this case searching or inserting an element will only need to decompress or compress a sub-block instead of a full block. Another drawback of compression techniques is that the number of stored elements may vary considerably from a block to another depending on how well the elements are compressed inside each block.

Applications

See also

• Binary search tree
• B-tree
• Divide-and-conquer algorithm

References

External links

Wikibooks has a book on the topic of: Algorithm Implementation/Trees/B+ tree

• B+ tree in Python, used to implement a list
• Dr. Monge's B+ Tree index notes
• Evaluating the performance of CSB+-trees on Mutithreaded Architectures
• Effect of node size on the performance of cache conscious B+-trees
• Fractal Prefetching B+-trees
• Towards pB+-trees in the field: implementations Choices and performance
• Cache-Conscious Index Structures for Main-Memory Databases
• Cache Oblivious B(+)-trees
• The Power of B-Trees: CouchDB B+ Tree Implementation
• B+ Tree Visualization
• B +-trees by Kerttu Pollari-Malmi
• Data Structures B-Trees and B+ Trees