快速动态命名集计算

Question

我有一个很长的复杂查询，其中包含很多计算和条件，但主要结构如下所示：

WITH
MEMBER [Id1] AS [Level].[Level1].CurrentMember.Member_Key
MEMBER [Id2] AS [Level].[Level2].CurrentMember.Member_Key
MEMBER [Level].[Level1].[FirstSet] AS NULL
MEMBER [Level].[Level1].[SecondSet] AS NULL
SET [Set 1] AS {some processed set members}
SET [Set 2] AS {some other processed set members}
SET [Common CrossJoin Set] AS [Level].[Level2].Members

MEMBER [Calculated Measure 1] AS
  IIF([Level].[Level].CurrentMember.Member_Key = 'FirstSet',
    SUM(existing [Set 1]),
    IIF([Level].[Level].CurrentMember.Member_Key = 'SecondSet',
      SUM(existing [Set 2]),
      SUM([Measures].[Measure1]) * 15
    )
  )

MEMBER [Calculated Measure 2] AS
IIF([Level].[Level].CurrentMember.Member_Key = 'FirstSet',
  SUM(existing [Set 1]),
  IIF([Level].[Level].CurrentMember.Member_Key = 'SecondSet',
    SUM(existing [Set 2]),
    SUM([Measures].[Measure2]) * 20
  )
)
SELECT 
  { [Id1], [Id2], [Calculated Measure 1], [Calculated Measure 2]} ON COLUMNS,
  { ([Common CrossJoin Set], [Level].[Level1].[FirstSet]),
    ([Common CrossJoin Set], [Level].[Level1].[SecondSet])
  } ON ROWS
FROM [Cube]

所以结果表如下所示：

║ ---------------║ ---------------║ Id1 ║ Id2 ║ 测量 1 ║ 测量 2 ║

║L2成员║L1.FirstSet成员║L2-1║L1-8║1║5║

║L2成员║L1.FirstSet成员║L2-2║L1-9║2║6║

║L2成员║L1.SecondSet成员║L2-3║L1-98║3║7║

║L2成员║L1.SecondSet成员║L2-4║L1-99║4║8║

结果正确，但查询速度很慢（>4 秒）。我的实际查询更大，并且包含很多这样的集合和度量，所以看起来问题出在现有功能和整体结构中，阻止了引擎执行内部优化。

这种方案是错误的，丑陋的，但我怎样才能重写它并更快地得到相同的结果呢？

Answer 1

我怀疑瓶颈是因为当您使用Iif 时，两个逻辑分支都不是NULL，所以您没有得到块模式计算：这是使用Iif 的更好方法：Iif(someBoolean, X, Null) 或Iif(someBoolean, Null, x) 但不幸的是，在你的情况下，你也不能有 null。

也许您可以尝试实现 Mosha 建议的这种模式来替换 Iif：

WITH 
MEMBER Measures.[Normalized Cost] AS [Measures].[Internet Standard Product Cost]
CELL CALCULATION ScopeEmulator 
  FOR '([Promotion].[Promotion Type].&[No Discount],measures.[Normalized Cost])' 
  AS [Measures].[Internet Freight Cost]+[Measures].[Internet Standard Product Cost]
MEMBER [Ship Date].[Date].RSum AS Sum([Ship Date].[Date].[Date].MEMBERS), SOLVE_ORDER=10
SELECT 
 [Promotion].[Promotion Type].[Promotion Type].MEMBERS on 0
 ,[Product].[Subcategory].[Subcategory].MEMBERS*[Customer].[State-Province].[State-Province].MEMBERS ON 1
FROM [Adventure Works]
WHERE ([Ship Date].[Date].RSum, Measures.[Normalized Cost])

这是来自这篇关于优化Iif的博客文章：http://sqlblog.com/blogs/mosha/archive/2007/01/28/performance-of-iif-function-in-mdx.aspx

所以看看你的一个计算 - 这个：

MEMBER [Calculated Measure 1] AS
  IIF([Level].[Level].CurrentMember.Member_Key = 'FirstSet',
    SUM(existing [Set 1]),
    IIF([Level].[Level].CurrentMember.Member_Key = 'SecondSet',
      SUM(existing [Set 2]),
      SUM([Measures].[Measure1]) * 15
    )
  )

我认为我们最初可以将其分解为：

MEMBER [Measures].[x] AS SUM(existing [Set 1])
MEMBER [Measures].[y] AS SUM(existing [Set 2])
MEMBER [Measures].[z] AS SUM([Measures].[Measure1]) * 15
MEMBER [Calculated Measure 1] AS
  IIF([Level].[Level].CurrentMember IS [Level].[Level].[Level].[FirstSet],
    [Measures].[x],
    IIF([Level].[Level].CurrentMember IS [Level].[Level].[Level].[SecondSet],
      [Measures].[y],
      [Measures].[z]
    )
  )  

现在尝试应用 Mosha 的模式（我之前没有尝试过，所以你需要相应地调整）

MEMBER [Measures].[z] AS SUM([Measures].[Measure1]) * 15
    MEMBER [Measures].[y] AS SUM(existing [Set 2])
    MEMBER [Measures].[x] AS SUM(existing [Set 1])
MEMBER [Calculated Measure 1 pre1] AS [Measures].[z]
CELL CALCULATION ScopeEmulator 
  FOR '([Level].[Level].[Level].[SecondSet],[Calculated Measure 1 pre1])' 
  AS [Measures].[y] 
MEMBER [Calculated Measure 1] AS [Calculated Measure 1 pre1]
CELL CALCULATION ScopeEmulator 
  FOR '([Level].[Level].[Level].[FirstSet],[Calculated Measure 1])' 
  AS [Measures].[x]