有一个通常的误解，觉得应该把选择性高的字段放在最前面，这通常只是针对一个字段的索引，对于组合索引，常常要把做等式运算的字段放在最前面，看看测试

USE AdventureWorks
GOCREATE TABLE demo1
( id INT identity(1,1)  PRIMARY KEY,gender char(1) NOT NULL ,age int NOT NULL,DESCRIPTION varchar(1000) default(replicate('a',1000))
)--填充数字辅助表
--DROP TABLE  #numSELECT row_number() over( order BY customerid ) AS  number
INTO #num
FROM adventureworks.sales.Individual--插入测试数据
INSERT demo1 (gender,age)
SELECT CASE WHEN number%2=0 THEN 'f'ELSE 'm'END,abs(checksum(newid()))%80
FROM #num

创建如下的索引：

CREATE INDEX ix_age_sex ON demo1(age,gender) include(description) WITH (online=on)

查询如下的语句：

DBCC FREEPROCCACHE
CHECKPOINT
DBCC DROPCLEANBUFFERSSELECT age,gender,description FROM demo1
WHERE gender='f'
AND age BETWEEN 30 AND 40

逻辑读为：

(1236 row(s) affected)
Table 'demo1'. Scan count 1, logical reads 350, physical reads 3, read-ahead reads 346, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.

现在删除原来的索引，创建低选择性在前的索引

DROP INDEX ix_age_sex ON demo1
CREATE INDEX ix_age_sex ON demo1(gender,age)INCLUDE (description) WITH (online=on)

在来查询相同的语句：

DBCC FREEPROCCACHE
CHECKPOINT
DBCC DROPCLEANBUFFERSSELECT age,gender,description FROM demo1
WHERE gender='f'
AND age BETWEEN 30 AND 40

逻辑读为：

(1236 row(s) affected)
Table 'demo1'. Scan count 1, logical reads 181, physical reads 3, read-ahead reads 178, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.

可以看到低选择性的字段放在前面逻辑读有350减少到了181，由此产生性能提升

总结：1：做等式运算的字段放在最前面，如果有多个等式运算，才要考虑等式运算字段的选择性