sql server 2005 优化方法

 

sql server 2005 优化方法

 

一分析实例级的等待

 

--可能存在的性能问题

（）IO等待；

（）存储过程和代码的重新编译；

（）tempdb数据库页可能是瓶颈；

 

 

--1 不用重新启动数据库实例，重置'sys.dm_os_wait_stats'

DBCC SQLPERF('sys.dm_os_wait_stats', CLEAR);

 

 

--2 查询累计值达到系统等待时间%的重量级等待

WITH Waits AS

(

  SELECT

    wait_type,

    wait_time_ms / 1000. AS wait_time_s,

    100. * wait_time_ms / SUM(wait_time_ms) OVER() AS pct,

    ROW_NUMBER() OVER(ORDER BY wait_time_ms DESC) AS rn

  FROM sys.dm_os_wait_stats

  WHERE wait_type NOT LIKE '%SLEEP%'

  -- filter out additional irrelevant waits

)

SELECT

  W1.wait_type,

  CAST(W1.wait_time_s AS DECIMAL(12, 2)) AS wait_time_s,

  CAST(W1.pct AS DECIMAL(12, 2)) AS pct,

  CAST(SUM(W2.pct) AS DECIMAL(12, 2)) AS running_pct

FROM Waits AS W1

  JOIN Waits AS W2

    ON W2.rn <= W1.rn

GROUP BY W1.rn, W1.wait_type, W1.wait_time_s, W1.pct

HAVING SUM(W2.pct) - W1.pct < 90 -- percentage threshold

ORDER BY W1.rn;

 

--3 找出高峰期

-- Create the WaitStats table

USE Performance;

GO

IF OBJECT_ID('dbo.WaitStats') IS NOT NULL

  DROP TABLE dbo.WaitStats;

GO

 

SELECT GETDATE() AS dt,

  wait_type, waiting_tasks_count, wait_time_ms,

  max_wait_time_ms, signal_wait_time_ms

INTO dbo.WaitStats

FROM sys.dm_os_wait_stats

WHERE 1 = 2;

 

ALTER TABLE dbo.WaitStats

  ADD CONSTRAINT PK_WaitStats PRIMARY KEY(dt, wait_type);

CREATE INDEX idx_type_dt ON dbo.WaitStats(wait_type, dt);

GO

 

-- Load waitstats data on regular intervals

INSERT INTO Performance.dbo.WaitStats

  SELECT GETDATE(),

    wait_type, waiting_tasks_count, wait_time_ms,

    max_wait_time_ms, signal_wait_time_ms

FROM sys.dm_os_wait_stats;

GO

 

-- Creation script for fn_interval_waits function

IF OBJECT_ID('dbo.fn_interval_waits') IS NOT NULL

  DROP FUNCTION dbo.fn_interval_waits;

GO

 

CREATE FUNCTION dbo.fn_interval_waits

  (@fromdt AS DATETIME, @todt AS DATETIME)

RETURNS TABLE

AS

 

RETURN

  WITH Waits AS

  (

    SELECT dt, wait_type, wait_time_ms,

      ROW_NUMBER() OVER(PARTITION BY wait_type

                        ORDER BY dt) AS rn

    FROM dbo.WaitStats

    WHERE dt >= @fromdt

      AND dt < @todt + 1

  )

  SELECT Prv.wait_type, Prv.dt AS start_time,

    CAST((Cur.wait_time_ms - Prv.wait_time_ms)

           / 1000. AS DECIMAL(12, 2)) AS interval_wait_s

  FROM Waits AS Cur

    JOIN Waits AS Prv

      ON Cur.wait_type = Prv.wait_type

      AND Cur.rn = Prv.rn + 1

      AND Prv.dt <= @todt;

GO

 

-- Return interval waits

SELECT wait_type, start_time, interval_wait_s

FROM dbo.fn_interval_waits('20060212', '20060215') AS F

ORDER BY SUM(interval_wait_s) OVER(PARTITION BY wait_type) DESC,

  wait_type, start_time;

GO

 

-- Prepare view for pivot table

IF OBJECT_ID('dbo.VIntervalWaits') IS NOT NULL

  DROP VIEW dbo.VIntervalWaits;

GO

 

CREATE VIEW dbo.VIntervalWaits

AS

 

SELECT wait_type, start_time, interval_wait_s

FROM dbo.fn_interval_waits('20060212', '20060215') AS F;

GO

 

--利用excel创建透视图分析

 

二联系等待和队列

 

--1 利用性能计数器监控cpu利用率，I/O队列，缓存命中率，内存计数器等。

 

--2 为服务器维护的每个性能计数器返回一行

SELECT

  object_name,

  counter_name,

  instance_name,

  cntr_value,

  cntr_type

FROM sys.dm_os_performance_counters;

 

 

三确定方案

 

--根据以上分析掌握的信息，确定下一步方案

 

四细化到数据库/文件级别

 

--分析IO信息

 WITH DBIO AS

(

  SELECT

    DB_NAME(IVFS.database_id) AS db,

    CASE WHEN MF.type = 1 THEN 'log' ELSE 'data' END AS file_type,

    SUM(IVFS.num_of_bytes_read + IVFS.num_of_bytes_written) AS io,

    SUM(IVFS.io_stall) AS io_stall

  FROM sys.dm_io_virtual_file_stats(NULL, NULL) AS IVFS

    JOIN sys.master_files AS MF

      ON IVFS.database_id = MF.database_id

      AND IVFS.file_id = MF.file_id

  GROUP BY DB_NAME(IVFS.database_id), MF.type

)

SELECT db, file_type,

  CAST(1. * io / (1024 * 1024) AS DECIMAL(12, 2)) AS io_mb,

  CAST(io_stall / 1000. AS DECIMAL(12, 2)) AS io_stall_s,

  CAST(100. * io_stall / SUM(io_stall) OVER()

       AS DECIMAL(10, 2)) AS io_stall_pct,

  ROW_NUMBER() OVER(ORDER BY io_stall DESC) AS rn

FROM DBIO

ORDER BY io_stall DESC;

 

五细化到进程级别

 

--1 注意事项：不要用SQL Server Profiler图形化跟踪；不要把跟踪数据写到数据库表；

不要把跟踪文件写到包含数据库文件的磁盘上；选择事件类和数据列，只跟踪需要的信息，

移除所有默认项和非必选项；列用筛选条件，如数据库ID；

 

 

--2 需要跟踪的事件类和字段

SP:Completed 

SP:StmtCompleted

RPC:Completed

SQL:StmtCompleted

 

TextData  Duration host application login

 

--3 用于创建跟踪的系统存储过程

--创建跟踪定义。新的跟踪将处于停止状态。

sp_trace_create

 

 

--创建用户定义事件

sp_trace_generateevent

 

--在跟踪中添加或删除事件或事件列。

sp_trace_setevent

 

--将筛选应用于跟踪。

sp_trace_setfilter

 

 

--修改指定跟踪的当前状态

--0 停止指定的跟踪。

--1 启动指定的跟踪。

--2 关闭指定的跟踪并从服务器中删除其定义。

sp_trace_setstatus

 

 

--以表格格式返回一或多个跟踪文件的内容

SELECT * INTO temp_trc

FROM fn_trace_gettable('c:/temp/my_trace.trc', default);