我们新建一个DataBase project，然后建立一个UserDefinedFunctions,Code像这样:
代码如下：

1: /// <summary>
/// SQLs the array.
/// </summary>
/// <param name="str">The STR.</param>
/// <param name="delimiter">The delimiter.</param>
/// <returns></returns>
/// 1/8/2010 2:41 PM author: v-pliu
[SqlFunction(Name = "CLR_Split",
FillRowMethodName = "FillRow",
TableDefinition = "id nvarchar(10)")]
public static IEnumerable SqlArray(SqlString str, SqlChars delimiter)
{
if (delimiter.Length == 0)
return new string[1] { str.Value };
return str.Value.Split(delimiter[0]);
}
/// <summary>
/// Fills the row.
/// </summary>
/// <param name="row">The row.</param>
/// <param name="str">The STR.</param>
/// 1/8/2010 2:41 PM author: v-pliu
public static void FillRow(object row, out SqlString str)
{
str = new SqlString((string)row);
}

然后Bulid,Deploy一切OK后，在SSMS中执行以下测试T-sql:
代码如下：

DECLARE @array VARCHAR(max)
SET @array = '39,15,93,68,64,43,90,58,39,9,26,26,89,47,91,57,98,16,55,9,63,29,69,16,41,76,34,60,68,64,61,53,32,30,11,72,57,63,36,43,22,14,60,38,24,5,66,26,26,21,22,99,55,18,7,10,46,76,27,88,9,29,89,75,48,72,94,59,35,19,0,35,79,11,87,49,68,30,91,35,9,7,34,47,41,61,98,13,22,1,26,80,35,48,34,92,24,85,90,51' SELECT id FROM dbo.CLR_Split(@array,',')

我们来看它的Client Statistic:

接着我们执行测试T-sql使用相同的array:
代码如下：

DECLARE @array VARCHAR(max)
SET @array = '39,15,93,68,64,43,90,58,39,9,26,26,89,47,91,57,98,16,55,9,63,29,69,16,41,76,34,60,68,64,61,53,32,30,11,72,57,63,36,43,22,14,60,38,24,5,66,26,26,21,22,99,55,18,7,10,46,76,27,88,9,29,89,75,48,72,94,59,35,19,0,35,79,11,87,49,68,30,91,35,9,7,34,47,41,61,98,13,22,1,26,80,35,48,34,92,24,85,90,51'
SELECT item FROM strToTable(@array,',')

CTE实现的Split function的Client statistic:

通过对比，你可以发现CLR的performance略高于CTE方式，原因在于CLR方式有Cache功能，并且把一个复杂的运算放到程序里比DataBase里更加高效。

您还可以参考：
Split string in SQL Server 2005+ CLR vs. T-SQL
Author:Petter Liu