[[email protected]

    [[email protected]

• 查看磁盤讀寫速度

    [[email protected] home]# time dd if=/dev/zero of=/home/4kb.1GBFILE bs=4k count=262144
    262144+0 records in
    262144+0 records out
    1073741824 bytes (1.1 GB) copied, 1.58541 s, 677 MB/s
    real 0m1.589s
    user 0m0.050s
    sys 0m1.533s

• 查看數據庫版本

    1> select @@version
    2> go
    --------------------------------------------------------------------------------------
    Adaptive Server Enterprise/15.7/EBF 21708 SMP SP110 /P/x86_64/Enterprise Linux/ase157sp11x/3546/64-bit/FBO/Fri Nov 8 05:39:38 2013
    (1 row affected)

二、數據準備

• 建立聚簇索引表（sybase主鍵默認為聚簇索引）

    USE DB_TASK
    GO
    CREATE TABLE T_TASKITEM_CI (
    C_BH char(32) primary key,
    C_BH_TASK char(32) null,
    C_BH_AJ varchar(32) null,
    N_AJBS numeric(15,0) null,
    C_AJLB varchar(6) null,
    N_JBFY int null,
    N_ZT int null,
    C_AH varchar(75) null
    )
    go

• 建立非聚簇索引表

    USE DB_TASK
    go
    CREATE TABLE T_TASKITEM_NCI (
    C_BH char(32) NOT NULL,
    C_BH_TASK char(32) null,
    C_BH_AJ varchar(32) null,
    N_AJBS numeric(15,0) null,
    C_AJLB varchar(6) null,
    N_JBFY int null,
    N_ZT int null,
    C_AH varchar(75) null
    )
    go
    CREATE UNIQUE INDEX PK_TASKITEM ON DB_TASK.dbo.T_TASKITEM_NCI (C_BH)
    go

• 構造數據
  構造一張同構的數據表T_TASKITEM_CC，使用如下SQL向該張表構造50W左右數據。

    SELECT newid ()
    , a.C_BH
    , 1 AS N_ZT
    , a.N_AJBS
    , a.N_JBFY
    , ‘5813b6d7ce8847d68b34daa956776659‘ AS C_BH_TASK
    , (CASE WHEN (a.N_YWLX = 20100) THEN ‘0201‘ WHEN (a.N_YWLX = 20200) THEN ‘0202‘ WHEN (a.N_YWLX = 20304) THEN ‘0207‘ WHEN (a.N_YWLX = 20501) THEN ‘0210‘ WHEN (a.N_YWLX = 20801) THEN ‘0224‘ WHEN (a.N_YWLX = 20601) THEN ‘0214‘ WHEN (a.N_YWLX = 20603) THEN ‘0216‘ WHEN (a.N_YWLX = 20602) THEN ‘0215‘ END) AS C_AJLB
    , a.C_AH
    FROM YWST..T_XS_AJ a

數據量為 501132 條

三、插入對比

• 非聚簇索引表

    1> insert into T_TASKITEM_NCI SELECT newid(),C_BH_TASK,C_BH_AJ,N_AJBS,C_AJLB,N_JBFY,N_ZT,C_AH FROM T_TASKITEM_CC
    2> GO
    Parse and Compile Time 0.
    Adaptive Server cpu time: 0 ms.
    Parse and Compile Time 0.
    Adaptive Server cpu time: 0 ms.
    Table: T_TASKITEM_NCI scan count 0, logical reads: (regular=2025588 apf=0 total=2025588), physical reads: (regular=0 apf=0 total=0), apf IOs used=0
    Table: T_TASKITEM_CC scan count 1, logical reads: (regular=10957 apf=27 total=10984), physical reads: (regular=0 apf=0 total=0), apf IOs used=0
    Total writes for this command: 3538
    Execution Time 97.
    Adaptive Server cpu time: 9688 ms. Adaptive Server elapsed time: 13381 ms.
    (501132 rows affected)

• 聚簇索引表

      1> insert into T_TASKITEM_CI SELECT newid(),C_BH_TASK,C_BH_AJ,N_AJBS,C_AJLB,N_JBFY,N_ZT,C_AH FROM T_TASKITEM_CC
      2> GO
      Parse and Compile Time 0.
      Adaptive Server cpu time: 0 ms.
      Parse and Compile Time 0.
      Adaptive Server cpu time: 0 ms.
      Table: T_TASKITEM_CI scan count 0, logical reads: (regular=6422447 apf=0 total=6422447), physical reads: (regular=0 apf=0 total=0), apf IOs used=0
      Table: T_TASKITEM_CC scan count 1, logical reads: (regular=10957 apf=27 total=10984), physical reads: (regular=0 apf=0 total=0), apf IOs used=0
      Total writes for this command: 11945
      Execution Time 176.
      Adaptive Server cpu time: 17350 ms. Adaptive Server elapsed time: 28206 ms.
      (501132 rows affected)

結論：插入同樣的數據量，非聚簇索引表比聚簇索引表時間上快一倍，IO減小2/3。

四、刪除對比

• 構造刪除數據

按照索引字段C_BH排序，獲取物理位置為於100行、200行…5000行的C_BH，將要刪除的編號分別存儲在T_DELETE_CI_BH 和 T_DELETE_NCI_BH表中。

    select C_BH,N_ORDER = identity(10) INTO T_ALL_CI_BH FROM T_TASKITEM_CI ORDER BY C_BH asc
    SELECT C_BH,N_ORDER INTO T_DELETE_CI_BH FROM T_ALL_CI_BH WHERE N_ORDER%100 = 0
    select C_BH,N_ORDER = identity(10) INTO T_ALL_NCI_BH FROM T_TASKITEM_NCI ORDER BY C_BH asc
    SELECT C_BH,N_ORDER INTO T_DELETE_NCI_BH FROM T_ALL_NCI_BH WHERE N_ORDER%100 = 0

• 聚簇索引表執行刪除

    1> DELETE FROM T_TASKITEM_CI where C_BH IN (SELECT C_BH FROM T_DELETE_CI_BH)
    2> go
    Parse and Compile Time 0.
    Adaptive Server cpu time: 0 ms.
    Parse and Compile Time 0.
    Adaptive Server cpu time: 0 ms.
    Parse and Compile Time 0.
    Adaptive Server cpu time: 0 ms.
    Table: T_TASKITEM_CI scan count 0, logical reads: (regular=20004 apf=0 total=20004), physical reads: (regular=0 apf=0 total=0), apf IOs used=0
    Table: T_DELETE_CI_BH scan count 1, logical reads: (regular=31 apf=0 total=31), physical reads: (regular=0 apf=0 total=0), apf IOs used=0
    Table: T_TASKITEM_CI scan count 5001, logical reads: (regular=15070 apf=0 total=15070), physical reads: (regular=0 apf=0 total=0), apf IOs used=0
    Total writes for this command: 241
    Execution Time 1.
    Adaptive Server cpu time: 128 ms. Adaptive Server elapsed time: 379 ms.
    (5001 rows affected)

• 非聚簇索引表執行刪除

    1> DELETE FROM T_TASKITEM_NCI where C_BH IN (SELECT C_BH FROM T_DELETE_NCI_BH)
    2> go
    Parse and Compile Time 0.
    Adaptive Server cpu time: 0 ms.
    Parse and Compile Time 0.
    Adaptive Server cpu time: 0 ms.
    Parse and Compile Time 0.
    Adaptive Server cpu time: 0 ms.
    Table: T_TASKITEM_NCI scan count 0, logical reads: (regular=20004 apf=0 total=20004), physical reads: (regular=0 apf=0 total=0), apf IOs used=0
    Table: T_DELETE_NCI_BH scan count 1, logical reads: (regular=31 apf=0 total=31), physical reads: (regular=0 apf=0 total=0), apf IOs used=0
    Table: T_TASKITEM_NCI scan count 5001, logical reads: (regular=15070 apf=0 total=15070), physical reads: (regular=0 apf=0 total=0), apf IOs used=0
    Total writes for this command: 242
    Execution Time 1.
    Adaptive Server cpu time: 128 ms. Adaptive Server elapsed time: 403 ms.
    (5001 rows affected)

結論：按照索引字段刪除，聚簇索引和非聚簇索引IO和效率一樣。

排查聚簇索引

聚簇索引表插入無序主鍵(GUID/UUID)數據時會造成額外的磁盤IO和時間消耗，采用無序主鍵(GUID/UUID)的項目設計上是禁止使用聚簇索引，那麽如何排查項目中非法使用聚簇索引的表呢？使用dba團隊出品的sp_dba_citable存儲過程檢索
核心代碼：

    use sybsystemprocs
    GO
    if object_id(‘sp_dba_citable‘) is not null
    drop procedure sp_dba_citable
    GO
    create procedure sp_dba_citable
    AS
    --查看聚簇索引表
    --add by wangzhen 2017-07-17
    begin
    declare @temp_sql varchar(500)
    declare @sql varchar(1000)
    declare @dbname varchar(100)
    declare dbname_cursor cursor for select name from master..sysdatabases
    create table #objectinfo (
    dbname varchar(100),
    objid int,
    tablename varchar(300),
    indexid int,
    indexname varchar(300),
    keycnt int,
    indextype varchar(100)
    )
    set @temp_sql = ‘insert into #objectinfo ‘
    　　　　+ ‘select ‘‘@dbname#‘‘ , ‘
    　　　　+ ‘　　obj.id , ‘
    　　　　+ ‘　　obj.name , ‘
    　　　　+ ‘　　ind.indid , ‘
    　　　　+ ‘　　ind.name , ‘
    　　　　+ ‘　　ind.keycnt , ‘
    　　　　+ ‘　　‘‘culster index‘‘ ‘
    　　　　+‘ from @dbname#..sysindexes ind left join @dbname#..sysobjects obj on ind.id = obj.id ‘
    　　　　+‘ where (ind.status2 & 512 = 512 or ind.indid = 1) and obj.type = ‘‘U‘‘ ‘
    open dbname_cursor
    while @@sqlstatus =0
    BEGIN
    　　FETCH dbname_cursor into @dbname
    　　set @sql = str_replace(@temp_sql,‘@dbname#‘,@dbname)
    　　EXECUTE(@sql)
    END
    close dbname_cursor
    select
    　　　　t.dbname as "庫名",
    　　　　t.objid as "對象ID",
    　　　　t.tablename as "表名",
    　　　　t.indexname as "索引名"
    from #objectinfo t where t.dbname not in (‘master‘,‘tempdb‘,‘sybsecurity‘,‘sybsystemdb‘,‘sybsystemprocs‘) group by t.dbname,t.objid,t.tablename,t.indexname,t.keycnt,t.indextype order by t.dbname asc,t.tablename asc
    end
    go

總結

在聚簇索引中,物理數據的存儲順序與索引相同,索引的最低級別包含實際的數據頁，在無序字段上（GUID/UUID）上使用聚簇索引插入大量數據會比非聚簇索引時間慢一倍，IO高三倍。其實，NP在設計之初已經規定業務表中不能定義物理主鍵（含聚簇索引），而應該定義邏輯主鍵（唯一約束+索引+不為空）。對於采用無序主鍵（GUID/UUID）的項目可以使用sp_dba_citable排查聚簇索引表！

聚簇索引對數據插入的影響