在创建主键或其它索引时,SQL Server总是自动将字段的顺序设置为升序排列;升序是默认设置,是为了保持与 SQL Server 早期版本的兼容性。建索引时索引列的顺序应该按照常用查询中的排序方式排序。 我们做个试验创建一个表,其中主键上的聚集索引按照id倒叙排列,然后分别倒叙顺序select数据,比较select的时间: +测试代码 if object_id('test_indexorder','U') is not null begin truncate table test_indexorder drop table test_indexorder end go create table test_indexorder ( id int identity(1,1) not null, name varchar(20) not null, content varchar(50) not null, co1 varchar(50), co2 varchar(50), co3 varchar(50), co4 varchar(50), co5 varchar(50), constraint pk_testorder primary key clustered( id desc ) ) go --insert 1000000 条数据 set nocount on; declare @t datetime; set @t = getdate(); DECLARE @cn int; set @cn = 1000000; while(@cn > 0) begin insert into test_indexorder(name,content,co1,co2,co3,co4,co5) VALUES( 'name' + cast(@cn as varchar(10)), cast(newid() as varchar(50)), cast(newid() as varchar(50)), cast(newid() as varchar(50)), cast(newid() as varchar(50)), cast(newid() as varchar(50)), cast(newid() as varchar(50))); set @cn = @cn -1; end print '插入时间(毫秒):'; print datediff(millisecond,@t,getdate()); set nocount off; GO checkpoint dbcc freeproccache dbcc dropcleanbuffers GO go set nocount on; declare @t datetime; set @t = getdate(); with t_rn as ( select *,rn = ROW_NUMBER() OVER (ORDER BY id desc) FROM test_indexorder ) SELECT id,name,content,co1,co2,co3,co4,co5 from t_rn WHERE rn between 19007 and 19057; print '查询时间(毫秒)' print datediff(millisecond,@t,getdate()) set @t = getdate(); with t_rn as ( select *,rn = ROW_NUMBER() OVER (ORDER BY id asc) FROM test_indexorder ) SELECT id,name,content,co1,co2,co3,co4,co5 from t_rn WHERE rn between 17007 and 17057; print '查询时间(毫秒)' print datediff(millisecond,@t,getdate()) set nocount off; 以下是查询时间结果 查询时间(毫秒) 393 查询时间(毫秒) 606 按照和索引相同顺序从100万条数据中取50条时需要393毫秒,相反顺序时需要606毫秒。造成的性能影响还是挺大的。 结论: 在建索引时要考虑常用查询的排序方式,在建主键时要特别注意,因为sql server会自动按照升序来建,这时候如果您的查询多数用主键列倒叙排列,记得要修改一下默认的设置。
我们做个试验创建一个表,其中主键上的聚集索引按照id倒叙排列,然后分别倒叙顺序select数据,比较select的时间:
+测试代码
if object_id('test_indexorder','U') is not null
begin
truncate table test_indexorder
drop table test_indexorder
end
go
create table test_indexorder
(
id int identity(1,1) not null,
name varchar(20) not null,
content varchar(50) not null,
co1 varchar(50),
co2 varchar(50),
co3 varchar(50),
co4 varchar(50),
co5 varchar(50),
constraint pk_testorder primary key clustered(
id desc
)
--insert 1000000 条数据
set nocount on;
declare @t datetime;
set @t = getdate();
DECLARE @cn int;
set @cn = 1000000;
while(@cn > 0)
insert into test_indexorder(name,content,co1,co2,co3,co4,co5)
VALUES(
'name' + cast(@cn as varchar(10)),
cast(newid() as varchar(50)),
cast(newid() as varchar(50)));
set @cn = @cn -1;
print '插入时间(毫秒):';
print datediff(millisecond,@t,getdate());
set nocount off;
GO
checkpoint
dbcc freeproccache
dbcc dropcleanbuffers
with t_rn as (
select *,rn = ROW_NUMBER() OVER (ORDER BY id desc) FROM test_indexorder
SELECT id,name,content,co1,co2,co3,co4,co5 from t_rn WHERE rn between 19007 and 19057;
print '查询时间(毫秒)'
print datediff(millisecond,@t,getdate())
select *,rn = ROW_NUMBER() OVER (ORDER BY id asc) FROM test_indexorder
SELECT id,name,content,co1,co2,co3,co4,co5 from t_rn WHERE rn between 17007 and 17057;
以下是查询时间结果
查询时间(毫秒)
393
606
按照和索引相同顺序从100万条数据中取50条时需要393毫秒,相反顺序时需要606毫秒。造成的性能影响还是挺大的。
结论:
在建索引时要考虑常用查询的排序方式,在建主键时要特别注意,因为sql server会自动按照升序来建,这时候如果您的查询多数用主键列倒叙排列,记得要修改一下默认的设置。