[转]SQL Server 索引基础知识(2)----聚集索引，非聚集索引

SQL Server 索引基础知识(2)----聚集索引，非聚集索引

[来自]http://blog.joycode.com/ghj/archive/2008/01/02/113291.aspx

由于需要给同事培训数据库的索引知识，就收集整理了这个系列的博客。发表在这里，也是对索引知识的一个总结回顾吧。通过总结，我发现自己以前很多很模糊的概念都清晰了很多。

不论是 聚集索引，还是非聚集索引，都是用B+树来实现的。我们在了解这两种索引之前，需要先了解B+树。如果你对B树不了解的话，建议参看以下几篇文章：

BTree,B-Tree,B+Tree,B*Tree都是什么
http://blog.csdn.net/manesking/archive/2007/02/09/1505979.aspx

B+ 树的结构图:

B+ 树的特点:

• 所有关键字都出现在叶子结点的链表中（稠密索引），且链表中的关键字恰好是有序的；
• 不可能在非叶子结点命中；
• 非叶子结点相当于是叶子结点的索引（稀疏索引），叶子结点相当于是存储（关键字）数据的数据层；

B+ 树中增加一个数据，或者删除一个数据，需要分多种情况处理，比较复杂，这里就不详述这个内容了。 

聚集索引（Clustered Index）

• 聚集索引的叶节点就是实际的数据页
• 在数据页中数据按照索引顺序存储
• 行的物理位置和行在索引中的位置是相同的
• 每个表只能有一个聚集索引
• 聚集索引的平均大小大约为表大小的5%左右

下面是两副简单描述聚集索引的示意图： 

在聚集索引中执行下面语句的的过程：

select * from table where firstName = 'Ota'

 

一个比较抽象点的聚集索引图示：

 

非聚集索引 （Unclustered Index）  

• 非聚集索引的页，不是数据，而是指向数据页的页。
• 若未指定索引类型，则默认为非聚集索引
• 叶节点页的次序和表的物理存储次序不同
• 每个表最多可以有249个非聚集索引
• 在非聚集索引创建之前创建聚集索引（否则会引发索引重建）

在非聚集索引中执行下面语句的的过程：

select * from employee where lname = 'Green'

一个比较抽象点的非聚集索引图示：

 

什么是 Bookmark Lookup

虽然SQL 2005 中已经不在提  Bookmark Lookup 了(换汤不换药)，但是我们的很多搜索都是用的这样的搜索过程，如下：
先在非聚集中找，然后再在聚集索引中找。

 

在 http://www.sqlskills.com/ 提供的一个例子中，就给我们演示了 Bookmark Lookup  比 Table Scan 慢的情况，例子的脚本如下：

USE CREDIT
go
-- These samples use the Credit database. You can download and restore the
-- credit database from here:
-- http://www.sqlskills.com/resources/conferences/CreditBackup80.zip
-- NOTE: This is a SQL Server 2000 backup and MANY examples will work on 
-- SQL Server 2000 in addition to SQL Server 2005.
-------------------------------------------------------------------------------
-- (1) Create two tables which are copies of charge:
-------------------------------------------------------------------------------
-- Create the HEAP
SELECT * INTO ChargeHeap FROM Charge
go
-- Create the CL Table
SELECT * INTO ChargeCL FROM Charge
go
CREATE CLUSTERED INDEX ChargeCL_CLInd ON ChargeCL (member_no, charge_no)
go
-------------------------------------------------------------------------------
-- (2) Add the same non-clustered indexes to BOTH of these tables:
-------------------------------------------------------------------------------
-- Create the NC index on the HEAP
CREATE INDEX ChargeHeap_NCInd ON ChargeHeap (Charge_no)
go
-- Create the NC index on the CL Table
CREATE INDEX ChargeCL_NCInd ON ChargeCL (Charge_no)
go
-------------------------------------------------------------------------------
-- (3) Begin to query these tables and see what kind of access and I/O returns
-------------------------------------------------------------------------------
-- Get ready for a bit of analysis:
SET STATISTICS IO ON
-- Turn Graphical Showplan ON (Ctrl+K)
-- First, a point query (also, see how a bookmark lookup looks in 2005)
SELECT * FROM ChargeHeap WHERE Charge_no = 12345
go
SELECT * FROM ChargeCL WHERE Charge_no = 12345
go
-- What if our query is less selective?
-- 1000 is .0625% of our data... (1,600,000 million rows)
SELECT * FROM ChargeHeap WHERE Charge_no < 1000
go
SELECT * FROM ChargeCL WHERE Charge_no < 1000
go
-- What if our query is less selective?
-- 16000 is 1% of our data... (1,600,000 million rows)
SELECT * FROM ChargeHeap WHERE Charge_no < 16000
go
SELECT * FROM ChargeCL WHERE Charge_no < 16000
go
-------------------------------------------------------------------------------
-- (4) What's the EXACT percentage where the bookmark lookup isn't worth it?
-------------------------------------------------------------------------------
-- What happens here: Table Scan or Bookmark lookup?
SELECT * FROM ChargeHeap WHERE Charge_no < 4000
go
SELECT * FROM ChargeCL WHERE Charge_no < 4000
go
-- What happens here: Table Scan or Bookmark lookup?
SELECT * FROM ChargeHeap WHERE Charge_no < 3000
go
SELECT * FROM ChargeCL WHERE Charge_no < 3000
go
-- And - you can narrow it down by trying the middle ground:
-- What happens here: Table Scan or Bookmark lookup?
SELECT * FROM ChargeHeap WHERE Charge_no < 3500
go
SELECT * FROM ChargeCL WHERE Charge_no < 3500
go
-- And again:
SELECT * FROM ChargeHeap WHERE Charge_no < 3250
go
SELECT * FROM ChargeCL WHERE Charge_no < 3250
go
-- And again:
SELECT * FROM ChargeHeap WHERE Charge_no < 3375
go
SELECT * FROM ChargeCL WHERE Charge_no < 3375
go
-- Don't worry, I won't make you go through it all :)
-- For the Heap Table (in THIS case), the cutoff is: 0.21%
SELECT * FROM ChargeHeap  WHERE Charge_no < 3383
go
SELECT * FROM ChargeHeap WHERE Charge_no < 3384
go
-- For the Clustered Table (in THIS case), the cut-off is: 0.21%
SELECT * FROM ChargeCL WHERE Charge_no < 3438
SELECT * FROM ChargeCL WHERE Charge_no < 3439
go

这个例子也就是 吴家震 在Teched 2007 上的那个演示例子。

小结：

这篇博客只是简单的用几个图表来介绍索引的实现方法：B+数， 聚集索引，非聚集索引，Bookmark Lookup 的信息而已。

参考资料：

表组织和索引组织
http://technet.microsoft.com/zh-cn/library/ms189051.aspx
http://technet.microsoft.com/en-us/library/ms189051.aspx

How Indexes Work
http://manuals.sybase.com/onlinebooks/group-asarc/asg1200e/aseperf/@Generic__BookTextView/3358

Bookmark Lookup
http://blogs.msdn.com/craigfr/archive/2006/06/30/652639.aspx 

Logical and Physical Operators Reference
http://msdn2.microsoft.com/en-us/library/ms191158.aspx