记一次sql server 性能调优，查询从20秒至2秒

一、需求

需求很简单，就是需要查询一个报表，只有1个表，数据量大约60万左右，但是中间有些逻辑。

先说明一下服务器配置情况：1核CPU、2GB内存、机械硬盘、Sqlserver 2008 R2、Windows Server2008 R2 SP1和阿里云的服务器，简单说就是阿里云最差的服务器。

1、原始表结构

非常简单的一张表，这次不讨论数据冗余和表字段的设计，如是否可以把Project和Baojian提出成一个表等等，这个是原始表结构，这个目前是没有办法改变的。

2、查询的sql语句为

select *from(
    select *,ROW_NUMBER() OVER (ORDER BY sc desc) as rank  
    from(
        select *,
                case when ( 40-(a.p*(case when a.p > 0 then 1 else -0.5 end)))<=30 
                        then 30 
                     else ( 40-(a.p*(case when a.p > 0 then 1 else -0.5 end))) 
                     end as sc   
        from (
        select * from (
                select a.ProjectNumber, a.ProjectName, a.BaojianNumber, a.BaojianName, a.ToubiaoPerson,
                    sum(UnitPrice) as sumPrice, 
                    b.price as avgPrice, 
                    ((sum(UnitPrice)-b.price)/nullif(b.price,0)*100) as p,
                    sum(case when UnitPrice>b.price then b.price else UnitPrice end )as pprice,
                    sum(case when UnitPrice>MaxPrice then 1 else 0 end ) as countChao
                from ToubiaoDetailTest1 a
                        join (
                            select ProjectNumber, ProjectName, BaojianNumber, BaojianName, avg(price) as price
                            from(
                                select * from(
                                    select ProjectNumber, ProjectName, BaojianNumber, BaojianName, ToubiaoPerson, 
                                        SUM(UnitPrice) as price,
                                        SUM(case when UnitPrice>MaxPrice then 1 else 0 end ) as countChao
                                    from ToubiaoDetailTest1  
                                    group BY ProjectNumber, ProjectName, BaojianNumber, BaojianName, ToubiaoPerson
                                ) tt 
                                where tt.countChao = 0
                            ) t
                            group by ProjectNumber, ProjectName, BaojianNumber, BaojianName