Addison Wesley SQL Performance Tuning Sep 2002 ISBN 0201791692

Therefore the rule-based optimizer makes a plan:find matching rows using the index on column2.. And the whole table can be scanned using two pagereads, whereas an index lookup would take

The non-standard EXPLAIN statement (see Table 17-1) is thevital way to find out what the optimizer has done We haven'tmentioned it up to now because this book's primary goal has

been to show what you can do before the fact But EXPLAIN is

the way to measure whether your estimates correspond to

DBMS reality In many shops, it's customary to get an EXPLAINfor every SQL statement before submitting it for execution That

is quite reasonable What's perhaps less reasonable is the

custom of trying out every transformation one can think of andsubmitting them all for explanation That is mere floundering.Understanding principles—in other words, estimating what's

rows.) That is a narrow search, and usually it's faster to

perform a narrow search with a B-tree rather than scan all rows

in the table Therefore the rule-based optimizer makes a plan:find matching rows using the index on column2

index for column2 Those facts change everything The equalsoperation will match on 60% of the rows, so it's not a narrowsearch And the whole table can be scanned using two pagereads, whereas an index lookup would take three page reads(one to lookup in the index, two more to fetch the data later).Therefore the cost-based optimizer makes a different plan: findmatching rows using a table scan

optimizers, as you can see from Table 17-1 The claims don'tmean much by themselves What's important is whether theoptimizer estimates cost correctly and how it acts on the

"Updates" Statistics for the Optimizer

Informix Yes SET EXPLAIN UPDATE STATISTICSIngres Yes EXECUTE QEP optimizedb utility

InterBase Yes SELECT … PLAN SET STATISTICS

Microsoft Yes EXPLAIN UPDATE STATISTICSMySQL No EXPLAIN ANALYZE TABLE

Oracle Yes EXPLAIN PLAN

Claims to be CBO column

This column is "Yes" if the DBMS's documentation makesthe claim that it operates with a cost-based optimizer

"Explains" the Access Plan column

This column shows the non-standard statement provided bythe DBMS so that you can examine the access plan theoptimizer will use to resolve an SQL statement For

example, if you want to know how Oracle will resolve a

specific SELECT statement, just execute an EXPLAIN PLANFOR statement for the SELECT

"Updates" Statistics for the Optimizer column

This column shows the non-standard statement or utilitythe DBMS provides so that you can update volatile

information, or statistics, for the optimizer For example, ifyour DBMS is MySQL and you've just added many rows to atable and want the optimizer to know about them, just

execute an ANALYZE TABLE statement for the table

This glossary contains only terms that specifically are used forSQL optimization For terms that apply to the subject of SQL ingeneral, consult the 1,000-term glossary on our Web site,

ourworld.compuserve.com/homepages/OCELOTSQL/glossary.htm

Before the definition there may be a "Used by" note For

example, "Used by: Microsoft, Sybase" indicates that Microsoftand Sybase authorities prefer the term and/or definition Thewords "Used by: this book only" indicate a temporary and non-standard term that exists only for this book's purposes

Second normal form table, a 1NF table that contains onlycolumns that are dependent upon the entire primary key

3NF

Third normal form table, a 2NF table whose non-key

columns are also mutually independent; that is, eachcolumn can be updated independently of all the rest

Application Programming Interface, the method by which aprogrammer writing an application program can make

requests of the operating system or another application

applet

A Java program that can be downloaded and executed by abrowser

B

B-tree

A structure for storing index keys; an ordered, hierarchical,paged assortment of index keys Some people say the "B"stands for "Balanced."

back compression

Making index keys shorter by throwing bytes away from theback

See also [compression]

See also [front compression]

See [B-tree]

BDB

Berkeley DB, an embedded database system that is bundledwith MySQL

Used by: Microsoft

A pointer in an index key If the data is in a clustered index,the bookmark is a clustered index key If the data is not in aclustered index, the bookmark is an RID

bytecode

An intermediate form of code in which executable Java

programs are represented Bytecode is higher level thanmachine code, but lower level than source code

cartesian explosion

The effect on the size of a cartesian join's product as thejoined tables grow in size During a join, the DBMS creates

a temporary table to hold the join result The temporarytable's size is the product of the sizes of the two originaltables, which means the processing time goes up

geometrically if Table1 and Table2 get bigger

cartesian join

The set of all ordered pairs {a, b} where a is a member ofset A and b is a member of set B In database terms, acartesian product joins all rows in Table1 with all rows inTable2 Thus if Table1 has the values {T_a1, T_b1} and Table2 has the values {T_a2, T_b2} then the cartesianproduct is {(T_a1, T_a2) (T_a1, T_b2) (T_b1, T_a2)(T_b1, T_b2)}.Cartesian products are useful for

explanation, but when we see an operation which "goescartesian," we usually criticize the optimizer Also known as

A sort order that considers 'SMITH' and 'Smith' to betwo different strings

CBO

Cost-based optimizer, an optimizer that uses volatile data(e.g., the row and column values that have been inserted)and an override (e.g., that the database contents are moreimportant than the fixed optimizing assumptions) to

determine the optimal query resolution plan A cost-basedoptimizer is a type of rule-based optimizer that has

additional volatile information available to it so that it canoverride a fixed assumption

CGI

Common Gateway Interface, a standard way for a Webserver to pass a Web user's request to an application

program and to receive data back to forward to the user

cluster

Used by: Microsoft, Sybase

A structure for storing data in a specific order; that is, an

permanently according to some column value, such as theprimary key

clustered index

An index that the DBMS uses to determine the order of datarows, according to values in one or more columns, calledthe cluster key With a strong-clustered index, the data

pages are the index's leaves and are thus always in order.With a weak-clustered index, data pages are separate fromindex leaf pages and the rows need not be 100% in order

See [compound index]

composite table

A table that contains column values derived from two ormore other tables

compound index

An index whose keys contain values derived from more thanone data column

compression

Making index keys shorter by throwing bytes away from thefront or from the back

Two transactions that have overlapping start or end times

To prevent concurrent transactions from interfering witheach other, the DBMS may arrange a lock

connection pooling

A facility that allows connections to a data source to bestored and reused A Java term

cursor

A marker that indicates the current position within a resultset

data source

A repository for storing data An ODBC/JDBC term

DBA

performs all activities related to maintaining a successfuldatabase environment

dbc

Database connection, an ODBC resource

DBMS

Database Management System, a program that lets one ormore computer users create and access the data in a

A condition that arises when two or more transactions arewaiting for one another to release locks

denormalize

Break the normalization rules deliberately, in an attempt togain speed or save space

{A,C,C,D,D} The number of distinct keys, that is, thenumber that a SELECT DISTINCT statement wouldreturn, is three: {A,C,D} Now recall from your arithmeticclasses that reciprocal = 1/N—that is, the reciprocal of anumber N is the number that would yield one when

multiplied by N The reciprocal of 3 is 1/3; therefore, thedensity for our example is 1/3 It's possible to get a

preliminary guess of the number of rows that column1 =

<literal> will return by multiplying the table's cardinality

WARNING: The following definitions of density from vendormanuals or other texts are imprecise or confusing: "theaverage number of rows which are duplicates" (Sybase);

"density is inversely proportional to index sensitivity"

(various)

dependence

A concept used in normalization If the value of column1uniquely determines the value of column2, then column2

is functionally dependent on column1 If the value of

column1 limits the possible values in column2 to a specificset, then column2 is set dependent on column1

dictionary sort

dictionary sort with tie-breaking

A dictionary sort with multiple passes to sort accentedcharacters differently from unaccented characters anduppercase letters differently from lowercase letters

Dirty Read

A problem arising with concurrent transactions The DirtyRead problem occurs when a transaction reads a row thathas been changed but not committed by another

transaction The result is that Transaction #2's work isbased on a change that never really happened You canavoid Dirty Read by using an isolation level of READ

The table that the DBMS examines first when evaluating ajoin or subquery expression

Enterprise Bean

A server component written in Java

env

A lock that may not coexist with any other lock on the sameobject

expanding update

A data-change statement that increases the size of a row

extent

A group of pages that are allocated together, as part of theinitial creation of an object or when an existing extent

A plan to process a subquery: make everything one level,that is, transform the query to a join, then process as ajoin

flattened (query)

See [flatten (a query)]

FPU

Floating Point Unit, a microprocessor that manipulatesnumbers faster than the basic microprocessor used by acomputer

[2] Used by: Informix What Informix users call

partitioning

See also [back compression]

functionally dependent

A concept used in normalization If the value of column1uniquely determines the value of column2, then column2

is functionally dependent on column1

G

granularity (of a lock)

The size of the locked area—database, file, table, page,row, or column

A method for producing a joined table Given two input

tables Table1 and Table2, processing is as follows:

(a) For each row in Table1, produce a hash Assign thehash to a hash bucket

(b) For each row in Table2, produce a hash Check if thehash is already in the hash bucket If it is: there's a join If

it is not: there's no join

A structure for storing data in an unstructured manner.When you add something to a heap, it goes wherever there

is free space, which probably means at the end Existingdata is not moved to make free space available for newdata

heap-organized table

See [heap]

histogram

Detailed information on the distribution of values over acolumn; information stored for the sake of the optimizer

expression Table1 LEFT JOIN Table2 the inner table

must be Table2 and for Table1 RIGHT JOIN Table2 theinner table must be Table1

L

latch

A low-level on-off mechanism that ensures two processes orthreads can't access the same object at the same time

See also [lock]

leaf (page of an index)

A page at the bottom level of a B-tree (the page at the toplevel is the root) Typically a leaf page contains pointers tothe data pages (if it's a non-clustered index) or to the dataitself (if it's a clustered index)

A method the DBMS uses to prevent concurrent

transactions from interfering with one another Physically, alock is one of three things: a latch, a mark on the wall, or aRAM record

locking level

See [granularity (of a lock)]

lock mode

The type of lock a DBMS has arranged for a transaction.Options are exclusive, shared, or update

Transaction #1's change never happened You can avoidLost Update by using an isolation level of READ

UNCOMMITTED or higher

LRU

Least-Recently-Used, an algorithm that replaces the pagethat hasn't been accessed for the longest time

M

mark on the wall

An ITL slot or mark put against a row by the DBMS Byputting a mark right beside the row accessed by a

materialize

See [materialization]

materialized view

A view whose rows take up space When you select from aview, the DBMS can elect to do one of two things: (a) it canget the rows from the original table, convert any derivedcolumns, and pass the results to the application or (b) itcan create a temporary table and put the rows from theoriginal table(s) into the temporary table, then select fromthe temporary copy The latter case results in a materializedview Materialization is often necessary when there is noone-to-one correspondence between the original table'srows and the view's rows (because there is a grouping) orwhen many tables are affected and concurrency would beharmed (because there is a join)

The process of finding a home for an expanding update.When a page overflows due to a data change that increasesthe length of a variable-length column, a row must be

Notice the for loop nested within a for loop

A problem arising with concurrent transactions The Non-repeatable Read by using an isolation level of REPEATABLEREAD or higher.

The process of designing a database so that its tables followthe rules specified by relational theory In practice, this

usually means that all database tables are in third normalform

process, according to rules based on relational theory In anormalized table, one set of columns is the primary key(which uniquely identifies a row of the table) and all othercolumns are functionally dependent upon the entire primarykey

Locking that assumes conflict is unlikely Generally, this

means avoiding locks and checking for conflict between twotransactions only after data changes have been made

outer table

The table in the outer loop of a nested-loop join When youwrite an SQL statement with an inner join, the outer table isdetermined by the DBMS based on its join strategy for that

of the join determines the outer table: for the join

expression Table1 LEFT JOIN Table2 the outer tablemust be Table1 and for Table1 RIGHT JOIN Table2 theouter table must be Table2

out-of-place update

Used by: Microsoft, Sybase A data change that causes arow to move

A fixed-size hopper that stores rows of data or index keys;

a minimal unit for disk I/O Depending on the DBMS, a page

is also called a data block, a block, a blocking unit, a controlinterval, or a row group

on separate disks Informix calls this fragmentation

A problem arising with concurrent transactions The

Phantom problem occurs when a transaction reads multiplerows twice; once before and once after another transactiondoes a data change that affects the search condition in thefirst transaction's reads The result is that Transaction #1gets a different (larger) result set back from its second

read You can avoid Phantoms by using an isolation level ofSERIALIZABLE

PL/SQL

Used by: Oracle

stored procedures

positioned delete

A DELETE statement that allows you to delete the row atthe current cursor position Syntax: DELETE … WHERE

CURRENT OF <cursor>

positioned update

An UPDATE statement that allows you to update the row atthe current cursor position Syntax: UPDATE … WHERE

CURRENT OF <cursor>

precompiler

A utility you use to "compile" SQL code before you compilethe host program, that is, a utility that converts SQL

statements in a host program to statements that a compilercan understand A remnant of embedded-SQL days; there is

no such thing as an SQL compiler

prepared statement

An SQL statement that has been parsed and planned, forexample, with ODBC's SQLPrepare function