Handout2 PDF

Free University of Bolzano–Database 2. Lecture II, 2003/2004 – A.
Artale (1)
Database 2
Lecture II
Alessandro Artale
Faculty of Computer Science – Free University of Bolzano
Room: 221
artale@inf.unibz.it
http://www.inf.unibz.it/ artale/
2003/2004 – First Semester

Free University of Bolzano–Database 2. Lecture II, 2003/2004 – A.Artale (2)
Summary of Lecture II
Indexing.
– Indexes on Sequential Files: Dense Vs. Sparse Indexes.
– Primary Indexes with Duplicate Keys.
– Secondary Indexes.
– Document Indexing.
– B-Tree Indexes.
Indexing
Indexing is the principal technique used to efficiently answering a given query.
An Index for a DB is like an Index in a book:

1. It is smaller that the book;
2. The words are in sorted order;
3. If we are looking for a particular topic we first search on the index, find
the pages where it is discussed, go to the actual pages in the book.
Example.
MovieStar(Name,Address,Gender,Birthdate)
SELECT *
FROM MovieStar
WHERE Name = ’Jim Carrey’;
All the blocks for the MovieStar relation should be inspected if there is no
index on Name.
Index
An Index is a data structure that facilitates the query answering process by

minimizing the number of disk accesses.
An index structure is usually defined on a single Attribute of a Relation,

called the Search Key;
An Index takes as input a Search Key value and returns the address of the
record(s) (block physical address offset of the record) holding that value.
Index structure: Search Key-Pointer pairs

Search Key Pointer to a data-file record
The Searck Key values stored in the Index are Sorted and a binary search can
be done on the Index.
Only a small part of the records of a relation have to be inspected: Ap-

propriate indexes can speed up query processing passing from minutes to
seconds.
Index Structures
Different data structures give rise to different indexes:
1. Indexes on Sequential Files (Primary Index);
2. Secondary Indexes on Unsorted Files;
3. B-Trees;
4. Hash Tables.
Evaluating Different Index Structures
No one technique is the best. Each has to evaluated w.r.t. the following criteria:
Access Type. Finding records either with a particular search key, or with the
search key falling in a given range.
Access Time. The time it takes to find item(s) using the index in question.
Insertion Time. The time to insert an item in the data file, as well as the
time to update the index.
Deletion Time. The time to delete the item from the data file (which include
the time to find the item), and the time to update the index.
Space Overhead. Additional space for the index.

Summary
Indexing
– B-Tree Indexes.
Indexes on Sequential Files
Index on Sequential File, also called Primary Index, when the Index is
associated to a Data File which is in turn sorted with respect to the search
key.
1. A Primary Index forces a sequential file organization on the Data File;
2. Since a Data File can have just one order there can be just one Primary
Index for Data File.
Usually used when the search key is also the primary key of the relation.
Usually, these indexes fit in main memory.
Indexes on sequential files can be:

1. Dense: One entry in the index file for every record in the data file;
2. Sparse: One entry in the index file for each block of the data file.
Dense Indexes
Every value of the search key has a representative in a Dense Index. The index
maintains the keys in the same order as in the data file.

Database System Implementation, H. Garcia-Molina, J. Ullman, and J. Widom, Prentice-Hall, 2000.

Queries with Dense Indexes
Algorithm for Lookup: Searching a data record with a given search key value.
Given a search key , the index is scanned and when is found the
associated pointer to the data file record is followed and the record (block
containing it) is read in main memory.
Dense indexes support also range queries: The minimum value is located
first, if needed, consecutive blocks are loaded in main memory until a search
key greater than the maximum value is found.
Query-answering using dense indexes is efficient:

1. Since the index is usually kept in main memory, just 1 disk I/O has to be
performed during lookup;
2. Since the index is sorted we can use binary search: If there are search

keys then at most steps are required to locate a given search key.

Sparse Indexes
Used when dense indexes are too large: A Sparse Index uses less space at the
expense of more time to find a record given a key.
A sparse index holds one key-pointer pair per data block, usually the first
record on the data block.


Queries with Sparse Indexes
Algorithm for Lookup.
Given a search key :

1. Search the sparse index for the greatest key to using binary search;
2. We retrieve the pointed block to main memory to look for the record with
search key (always using binary search).
With respect to dense indexes we need to start two different binary searches:
the first on the sparse index, and the second on the retrieved data block.
Still disk I/O for lookup.

In conclusion, a Sparse Index is more efficient in space at the cost of a worst

computing time in Main Memory.
Primary Dense Index: Example
Example of a Primary Dense Index with Search Key=Account#.
Account Branch Balance

A-101
A-101 Downtown 500
A-102
A-102 Perryridge 400
A-110
A-110 Downtown 600
A-201
A-215
A-215 Mianus 700
A-217
A-217 Brighton 750
A-218
A-222
A-222 Redwood 700
A-305
A-305 Round Hill 350
Primary Sparse Index: Example
Example of a Primary Sparse Index with Search Key=Account#.

A-101 Downtown 500
A-110 Downtown 600
A-101
A-201
A-215 Mianus 700
A-218
A-217 Brighton 750

A-222 Redwood 700
Summary
Indexing
– B-Tree Indexes.
Primary Indexes with Duplicate Keys
Indexes for non key attributes:

More than one record with the same search key.
As usual, the data file should be sorted w.r.t the search key to speak of
primary indexes.
Techniques for dense indexes:

1. One entry for each record in the data file: Duplicate key-pointer pairs
(not used);
2. Just a single entry for each record in the data file with search key – no
duplicate key-pointer pairs: Pointer to the first record with search key
(more efficient).
Dense Index with Duplicate Search Keys
Single Entry Index
Lookup. Find the search key on the index, read the pointed disk block, possibly
read successive blocks.


Primary Dense Index with Duplicates: Example
Example of a Primary Dense Index with Duplicates with Search Key=Branch.

A-217 Brighton 750
Brighton A-101 Downtown 500
Downtown A-110 Downtown 600
Mianus A-215 Mianus 700

Perryridge A-102 Perryridge 400
Redwood A-201 Perryridge 900
Round Hill A-218 Perryridge 700
A-222 Redwood 700
Analysis of Primary indexes
Advantages.
Efficient access of tuples with a given search key.
– Very few blocks should be read (also in case of duplicate keys);
– Range Queries – looking for search key values in a certain range – are
answered efficiently.
Analysis of Primary indexes (cont.)
Disadvantages.
Expensive maintenance of the physical records storage to maintain the sorted
order.
– Technique used for insertion based on Overflow Blocks.
1. If there is space in the block insert the new record there in the right
place;
2. Otherwise, insert the new record in an Overflow Blocks. In order
to maintain the order, records are linked by means of pointers: The
pointer in each record points to the next record in search-key order.
– In general, performance degrades as far as the relation grows. The file is
reorganized when the system load is low.
– An optimal solution is to implement primary indexes as B-Tree structures
(presented soon).
Insertion in Sequential Files: Example
A-215 Mianus 700 A-306 Mianus 650

Overflow Block
Summary
Indexing
– B-Tree Indexes.
Secondary Indexes
A primary index is an index on a file sorted w.r.t. the search key. Then a
primary index “controls” the storage of records in the data file.
Indexes on Sequential files and Hash Tables are examples of primary indexes.
Since a file can have at most one physical order then it can have at most one
primary index.
Secondary Indexes facilitate query-answering on attributes other than

primary keys – or, more generally, on non-ordering attributes.
A file can have several secondary indexes.
Secondary indexes do not determine the placement of records in the data file.
Secondary Index: An Example
Let us consider the MovieStar relation:

MovieStar(Name,Address,Gender,Birthdate)
and a query involving the non-key Birthdate attribute:
SELECT Name, Address

FROM MovieStar
WHERE Birthdate = ’1975-01-01’;
A secondary index on the MovieStar relation w.r.t. the Birthdate attribute

would reduce the answering time.
Structure of Secondary Indexes
Secondary Indexes are always Dense:

Sparse secondary indexes make no sense!
Secondary indexes are sorted w.r.t. the search key Binary search.
The Data File IS NOT sorted w.r.t. the Secondary Index Search Key!
More than one data block may be needed for a given search key in general
more disk I/O to answer queries:
– Secondary Indexes are less efficient than Primary Indexes.
Secondary Indexes: An Example
The example shows that 3 data blocks (i.e., 3 disk I/O) are needed to retrieve all
the tuples with search key

using the Secondary Index.


Indirect Buckets
To avoid repeating keys in secondary index, use a level of indirection, called
Buckets.
The index maintains only one key-pointer pair for each search key : The
pointer for goes to a position in the bucket which contains pointers to
records with search key till the next position pointed by the index.


Summary
Indexing
– B-Tree Indexes.
Document Retrieval and Inverted Indexes
Problem: Given a set of text documents we need to retrieve that ones where a
particular word(s) occurs.
Given the success of the Web this has become an urgent database problem.
A document is thought as a tuple in the relation Doc(ID,cat,dog, ),

with one attribute for each possible word.
Each attribute has a Boolean value, eg, the value of cat is TRUE if and only
if the word cat appears in the document.
An Inverted Index is a form of secondary index with indirect bucket

containing – as search keys – all the attribute names of the Doc relation.
Pointers are stored in a bucket file and consider only the TRUE occurrences
of a search key.
Inverted Indexes: An Example


Additional Information in Buckets
Buckets can be extended to include “Type” (e.g., specify whether the word
appears in the title, abstract or body), “Position” of word, etc.


Summary
Indexing
– B-Tree Indexes.
B-Trees
A B-Tree is a multilevel index with a Tree structure;
When used as primary index (i.e., on a sorted file) maintains efficiency

against insertion and deletion of records avoiding file reorganization (the
main disadvantage of index on sequential file);
Also used to index very-large relations when single-level indexes don’t fit in
main memory;
Commercial systems (DB 2, ORACLE) implement indexes with B-Trees;

In the following we will present the structure of so called -Tree – the B
stands for Balanced Tree.
B-Trees (cont.)
B-tree is usually a 3 levels tree: the root, an intermediate level, the leaves.
All the leaves are at the same level Balanced Tree.
The size of each node of the B-tree is equal to a disk block. All nodes have
the same format: keys and pointers key-pointer pairs plus

extra pointer.

To keys To keys
To keys To keys

Example. Let a block be bytes, a search key be an integer of bytes, and

a pointer be bytes. If there is no additional header in the block then is the

largest integer s.t. .

B-Tree: An Example

Data file where search-keys are all the prime numbers from to .

All the keys appear once (in case of a dense index), and in sorted order at the
leaves.
A pointer points to either a file record (in case of a primary index structure)
or to a bucket of pointers (in case of a secondary index structure).
Leaves

To next leaf
in sequence
To record To record To record

with key with key with key

One pointer to next leaf—used to chain leaf nodes in search-key order for
efficient sequential processing;

At least (round down) key-pointer pairs pointing to either records of

the data file (as shown in the Figure) or to a bucket of pointers.

Interior Nodes

To keys To keys
To keys To keys

All pointers can be used to point to B-tree nodes of the inferior level;

(round up) pointers must be used, and one more pointer than

At least

key is used.
– Exception: the root may have only 2 children, then one key and two
pointers, regardless of how large is.

B-Trees as Indexes
B-Trees are useful for various types of indexes:
1. The search key is a primary or candidate key (i.e., no duplicates) of the data
file and the B-Tree is a dense index with a key-pointer pair in a leaf for
every record in the data file. The data file may or may not be sorted by
primary key (primary or secondary index).
2. The search key is not a key (i.e., duplicate values) and the data file is sorted
by this attribute. The B-Tree is a dense primary index with no duplicate
key-pointer pairs: just a single entry for each record in the data file with
search key , and pointers to the first record with search key .
3. The data file is sorted by search-key, and the B-Tree is a sparse primary
index with a key-pointer pair for each data block of the data file.
4. The search key is not a key (i.e., duplicate values) and the data file is NOT
sorted. The B-Tree is a secondary index with indirect bucket: No
duplicate key-pointer pairs, just a single entry for each record in the data file
with a given search key, and pointers to a bucket of pointers.
Lookup in B-trees
Problem: Given a B-tree (dense) index, find a record with search key .
Recursive search, starting at the root and ending at a leaf:
1. If we are at a leaf then if is among the keys of the leaf follow the associated
pointer to the data file, else fail.
2. If we are at an interior node (included the root) with keys ,

then if then go to the first child, if then go to the

second child, and so on.
Note: B-Trees are useful for queries in which a range of values are asked for:
Range Queries.
B-Tree Updates
Insertion and Deletion are more complicated that lookup. It may be necessary to
either:
1. Split a node that becomes too large as the result of an insertion;
2. Merge nodes (i.e., combine nodes) if a node becomes too small as the result
of a deletion.
B-Tree Insertion
Algorithm for inserting a new search key in a B-Tree.
1. Start a search for the key being inserted. If there is room for another
key-pointer at the reached leaf, insert there;
2. If there is no room in the leaf, split the leaf in two and divide the keys
between the two new nodes (each node is at least half full);
3. The splitting of nodes implies that a new key-pointer pair has to be inserted
at the level above. If necessary the parent node will be split and we proceed
recursively up the tree (including the root).
B-Tree Insertion: Splitting Leaves

Let be a leaf whose capacity is keys, and we need to insert the

key-pointer pair.

1. Create a new sibling node , to the right of ;

2. The first key-pointer pairs remain with , while the other move to

.
3. The first key of the new node is also inserted at the parent node.

Note: At least key-pointer pairs for both of the splitted nodes.

B-Tree Insertion: Splitting Interior Nodes

Let be an interior node whose capacity is keys and pointers, and

has been assigned the new pointer because of a node splitting at the

inferior level.

1. Create a new sibling node , to the right of ;

2. Leave at the first pointers, and move the other to

;

3. The first keys stay with , while the last keys move to . Since

there are keys there is one key in the middle (say it ) that doesn’t

go with neither nor , but:

is reachable via the first of ’s children;

is used by the common parent of and to distinguish the search
between those two nodes.

Note: At least pointers for both of the splitted nodes.

Example: B-Tree Insertion of the key 40

Leaf of insertion
Example: Splitting the Leaf for Inserting Key 40

Example: Splitting Interior Nodes

B-Tree Deletion
Algorithm for deleting a search key in a B-Tree.
1. Start a search for the key being deleted;
2. Delete the record from the data file and the key-pointer pair from the leaf of
the B-tree;
3. If the lower limit of keys and pointers on a leaf is violated then two cases are
possible:
(a) Look for an adjacent sibling that is above lower limit and “steal” a
key-pointer pair from that leaf, keeping the order of keys intact. Make
sure keys for the parent are adjusted to reflect the new situation.
(b) Hard case: no adjacent sibling can provide an extra key. Then there must
be two adjacent siblings leaves, one at minimum, one below minimum
capacity. Just enough to merge nodes deleting one of them. Keys at
the parent should be adjusted, and then delete a key and a pointer. If
the parent is below the minimum capacity then we recursively apply the
deletion algorithm at the parent.
Example: B-Tree Deletion of the Key 7

Example: B-Tree Deletion of the Key 7 (cont.)

Efficiency of B-Trees
B-Trees allow lookup, insertion and deletion of records of very large relations
using few disk I/O.
When the capacity of B-Tree nodes is reasonably large ( ) splitting

and merging of nodes is rare.
3 levels are typical: let a block be bytes, a search key be an integer

of bytes, and a pointer be bytes . Suppose that a node

is occupied midway between the minimum ( ) and the maximum, then

each node has pointers Root children leaves

million pointers to data file records.

If the root is kept in main memory lookup requires disk I/O for traversing

the tree and disk I/O for accessing the record, if also second level in main

memory a single disk I/O is sufficient for traversing the tree.

Efficiency of B-Trees (cont.)
B-Tree maintains its efficiency against relation updates.
A relation is physically stored into a B-Tree.
The actual records are stored in the leaf level of the B-Tree.
Insertion and deletion can cause either node splitting or merging – i.e., no
need for overflow blocks.
Efficiency of B-Trees: Relation Storage Example

A-218
A-110 A-215 A-305
(A-101,Downtown,500) (A-102,Perryridge,400) (A-218,..) (A-222,..)
(A-110,Downtown,600) (A-201,Perryridge,900) (A-305,..) (A-306,..)
(A-215,Mianus,700) (A-217,Brighton,750)
Summary of Lecture II
Indexing
– B-Tree Indexes.

Handout2 PDF

Uploaded by

Copyright:

Available Formats

Handout2 PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Handout2 PDF

Uploaded by

Copyright:

Available Formats

Free University of Bolzano–Database 2. Lecture II, 2003/2004 – A.

2003/2004 – First Semester

Indexing is the principal technique used to efficiently answering a given query.

An Index for a DB is like an Index in a book:

An Index is a data structure that facilitates the query answering process by

An index structure is usually defined on a single Attribute of a Relation,

Index structure: Search Key-Pointer pairs

Only a small part of the records of a relation have to be inspected: Ap-

Different data structures give rise to different indexes:

1. Indexes on Sequential Files (Primary Index);

2. Secondary Indexes on Unsorted Files;

Evaluating Different Index Structures

Space Overhead. Additional space for the index.

Indexes on Sequential Files

Usually, these indexes fit in main memory.

Indexes on sequential files can be:

Database System Implementation, H. Garcia-Molina, J. Ullman, and J. Widom, Prentice-Hall, 2000.

Queries with Dense Indexes

Query-answering using dense indexes is efficient:

Database System Implementation, H. Garcia-Molina, J. Ullman, and J. Widom, Prentice-Hall, 2000.

Queries with Sparse Indexes

Algorithm for Lookup.

Given a search key :

Still disk I/O for lookup.

In conclusion, a Sparse Index is more efficient in space at the cost of a worst

Example of a Primary Dense Index with Search Key=Account#.

Account Branch Balance

Example of a Primary Sparse Index with Search Key=Account#.

Account Branch Balance

A-218 Perryridge 700

Primary Indexes with Duplicate Keys

Indexes for non key attributes:

Techniques for dense indexes:

Database System Implementation, H. Garcia-Molina, J. Ullman, and J. Widom, Prentice-Hall, 2000.

Example of a Primary Dense Index with Duplicates with Search Key=Branch.

Account Branch Balance

Brighton A-101 Downtown 500

Downtown A-110 Downtown 600

Mianus A-215 Mianus 700

Analysis of Primary indexes

Analysis of Primary indexes (cont.)

Insertion in Sequential Files: Example

A-215 Mianus 700 A-306 Mianus 650

Secondary Indexes facilitate query-answering on attributes other than

A file can have several secondary indexes.

Secondary Index: An Example

Let us consider the MovieStar relation:

SELECT Name, Address

A secondary index on the MovieStar relation w.r.t. the Birthdate attribute

Structure of Secondary Indexes

Secondary Indexes are always Dense:

Secondary Indexes: An Example

Database System Implementation, H. Garcia-Molina, J. Ullman, and J. Widom, Prentice-Hall, 2000.

Database System Implementation, H. Garcia-Molina, J. Ullman, and J. Widom, Prentice-Hall, 2000.

Document Retrieval and Inverted Indexes

A document is thought as a tuple in the relation Doc(ID,cat,dog, ),

An Inverted Index is a form of secondary index with indirect bucket

Inverted Indexes: An Example

Database System Implementation, H. Garcia-Molina, J. Ullman, and J. Widom, Prentice-Hall, 2000.