Question

我想从字段中找到最高的AutoIncremented值。（在我可以使用@@SCOPE_IDENTITY等插入后，它不被取出）这两个查询中的哪一个会运行得更快或性能更好。 Id是autoincrement的主键和Table1字段。这适用于Sql Server 2005.

SELECT MAX(Id) FROM Table1

SELECT TOP 1 Id FROM Table1 ORDER BY Id DESC

[编辑]
在这种情况下，是Id是我已定义聚簇索引的字段如果索引是ID DESC那么什么..
是的，如果知道如何影响性能会很好 1. Id是聚簇索引+主键 2. Id是聚簇索引而不是主键 3. Id是非聚集索引ASC +主键 4. Id是非聚集索引ASC而不是主键 5. Id是非聚集索引DESC +主键 6. Id是非聚集索引DESC而不是主键 7. Id只是AutoIncrement

希望它不是一个很高的命令！

Answer 1

如果存在聚集索引，则两个查询之间的性能几乎没有差异。

这是因为它们都会执行聚集索引扫描，它将承担100％的查询费用。

对没有索引的列执行两个查询会导致在两个执行计划中使用3个运算符。

Top子句使用Sort运算符，Max函数使用Stream Aggregate运算符。

当没有索引时，MAX（）函数提供更好的性能。

可以找到概念证明，并且可以找到测试场景的完整演练here

Performance Comparison Top 1 Verses MAX() Funciton

Answer 2

没人提到IDENT_CURRENT（'Table1'） - 将它们全部吹走 - 当然它只适用于标识列，但是问题......

Answer 3

从理论上讲，他们会使用相同的计划并且几乎同时运行。

在实践中，

SELECT TOP 1 Id FROM Table1 ORDER BY Id DESC

更可能使用PRIMARY KEY INDEX。

此外，如果您决定选择其他列以及id，则此广告更具扩展性。

MAX()上的实际计划说：

SELECT <- AGGREGATE <- TOP <- CLUSTERED INDEX SCAN

，而TOP 1的计划说：

SELECT <- TOP <- CLUSTERED INDEX SCAN

，我。即aggregate被省略。

聚合实际上不会在这里做任何事情，因为只有一行。

P上。 S。注意到@Mehrdad Afshari和@John Sansom，在非索引字段MAX上稍快一些（当然不是优化程序所说的20次）：< / p>

-- 18,874,368 rows

SET LANGUAGE ENGLISH
SET STATISTICS TIME ON
SET STATISTICS IO ON
PRINT 'MAX'
SELECT MAX(id) FROM master
PRINT 'TOP 1'
SELECT TOP 1 id FROM master ORDER BY id DESC
PRINT 'MAX'
SELECT MAX(id) FROM master
PRINT 'TOP 1'
SELECT TOP 1 id FROM master ORDER BY id DESC
PRINT 'MAX'
SELECT MAX(id) FROM master
PRINT 'TOP 1'
SELECT TOP 1 id FROM master ORDER BY id DESC
PRINT 'MAX'
SELECT MAX(id) FROM master
PRINT 'TOP 1'
SELECT TOP 1 id FROM master ORDER BY id DESC
PRINT 'MAX'
SELECT MAX(id) FROM master
PRINT 'TOP 1'
SELECT TOP 1 id FROM master ORDER BY id DESC

Changed language setting to us_english.

SQL Server Execution Times:
   CPU time = 0 ms,  elapsed time = 1 ms.

SQL Server Execution Times:
   CPU time = 0 ms,  elapsed time = 1 ms.

SQL Server Execution Times:
   CPU time = 0 ms,  elapsed time = 1 ms.
MAX

SQL Server Execution Times:
   CPU time = 0 ms,  elapsed time = 20 ms.

(строк обработано: 1)
Table 'master'. Scan count 3, logical reads 32655, physical reads 0, read-ahead reads 447, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.

SQL Server Execution Times:
   CPU time = 5452 ms,  elapsed time = 2766 ms.
TOP 1

SQL Server Execution Times:
   CPU time = 0 ms,  elapsed time = 1 ms.

(строк обработано: 1)
Table 'master'. Scan count 3, logical reads 32655, physical reads 0, read-ahead reads 2, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.

SQL Server Execution Times:
   CPU time = 6813 ms,  elapsed time = 3449 ms.
MAX

SQL Server Execution Times:
   CPU time = 0 ms,  elapsed time = 1 ms.

(строк обработано: 1)
Table 'master'. Scan count 3, logical reads 32655, physical reads 0, read-ahead reads 44, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.

SQL Server Execution Times:
   CPU time = 5359 ms,  elapsed time = 2714 ms.
TOP 1

SQL Server Execution Times:
   CPU time = 0 ms,  elapsed time = 1 ms.

(строк обработано: 1)
Table 'master'. Scan count 3, logical reads 32655, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.

SQL Server Execution Times:
   CPU time = 6766 ms,  elapsed time = 3379 ms.
MAX

SQL Server Execution Times:
   CPU time = 0 ms,  elapsed time = 1 ms.

(строк обработано: 1)
Table 'master'. Scan count 3, logical reads 32655, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.

SQL Server Execution Times:
   CPU time = 5406 ms,  elapsed time = 2726 ms.
TOP 1

SQL Server Execution Times:
   CPU time = 0 ms,  elapsed time = 1 ms.

(строк обработано: 1)
Table 'master'. Scan count 3, logical reads 32655, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.

SQL Server Execution Times:
   CPU time = 6780 ms,  elapsed time = 3415 ms.
MAX

SQL Server Execution Times:
   CPU time = 0 ms,  elapsed time = 1 ms.

(строк обработано: 1)
Table 'master'. Scan count 3, logical reads 32655, physical reads 0, read-ahead reads 85, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.

SQL Server Execution Times:
   CPU time = 5392 ms,  elapsed time = 2709 ms.
TOP 1

SQL Server Execution Times:
   CPU time = 0 ms,  elapsed time = 1 ms.

(строк обработано: 1)
Table 'master'. Scan count 3, logical reads 32655, physical reads 0, read-ahead reads 10, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.

SQL Server Execution Times:
   CPU time = 6766 ms,  elapsed time = 3387 ms.
MAX

SQL Server Execution Times:
   CPU time = 0 ms,  elapsed time = 1 ms.

(строк обработано: 1)
Table 'master'. Scan count 3, logical reads 32655, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.

SQL Server Execution Times:
   CPU time = 5374 ms,  elapsed time = 2708 ms.
TOP 1

SQL Server Execution Times:
   CPU time = 0 ms,  elapsed time = 1 ms.

(строк обработано: 1)
Table 'master'. Scan count 3, logical reads 32655, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.

SQL Server Execution Times:
   CPU time = 6797 ms,  elapsed time = 3494 ms.

Answer 4

只需比较执行计划，您就会看到（在编辑查询时按下Management Studio中的Ctrl+M）。我的猜测是，如果Id列上有（聚集的）索引，这些查询同样具有高效性。

然而，这整体上是非常糟糕的想法。

Answer 5

MAX通常更快。

如果存在，两个查询都将使用列上的索引。

如果列中不存在索引，则TOP 1查询将使用Top N Sort运算符排序表而不是流聚合，这会让它变慢。

MAX也提供了更好的可读性。

Side Note ：虽然MAX将在索引案例中的执行计划中使用流聚合运算符，但它没有任何特定成本，因为它只处理单行（ Actual Rows = 1）。您可以通过在一个批处理中运行查询来比较查询，并查看相对成本。在索引的情况下，两个查询都将花费50％。我在一张大约7000行的表格上测试了非索引的情况，与MAX相比，TOP的成本为65％，费用为35％。

Answer 6

我刚刚测试了您针对典型数据集提供的两个SQL语句：

SELECT MAX(Id) FROM Table1

SELECT TOP 1 Id FROM Table1 ORDER BY Id DESC

SELECT TOP 1 Id FROM Table1 ORDER BY Id DESC稍快一点，因为它在执行计划中有最后一步。以下是每个查询执行的执行计划：

SELECT MAX（Id）FROM Table1

聚集索引扫描＆gt;＆gt;顶部＆gt;＆gt;流聚合＆gt;＆gt;选择

SELECT ONE 1 Id FROM Table1 ORDER BY Id DESC

聚集索引扫描＆gt;＆gt;顶部＆gt;＆gt;选择

Answer 7

是的，在这种情况下，Id是字段   我已经定义了群集   指数。如果索引是ID DESC那么   什么..是的，它会很好   知道性能如何   如果
受到影响

Id是聚集索引+主键。

Id是聚集索引，而不是主键。

Id是非聚集索引ASC +主键。

Id是非聚集索引ASC，而不是主键。

Id是非聚集索引DESC +主键。

Id是非聚集索引DESC而不是主键。

Id只是AutoIncrement

对于案例1和案例2，两者都将执行返回单个记录的聚簇索引扫描。两个查询之间没有IO差异。

对于案例3,4,5和6，两者都将执行返回单个记录的索引扫描。两个查询之间没有IO差异。

对于案例7，两者都将执行表扫描。 IO成本没有差异。

总结：案例1-6取胜！如果你在案例7中，那么你已经从IO的角度失去了。

您可以使用SQL的查询分析器来测量IO。在查询之前运行此命令。

SET STATISTICS IO ON

对于自动增量字段：MAX（ID）vs TOP 1 ID ORDER BY ID DESC

7 个答案: