91超碰碰碰碰久久久久久综合_超碰av人澡人澡人澡人澡人掠_国产黄大片在线观看画质优化_txt小说免费全本

溫馨提示×

溫馨提示×

您好,登錄后才能下訂單哦!

密碼登錄×
登錄注冊×
其他方式登錄
點擊 登錄注冊 即表示同意《億速云用戶服務條款》

部署statspack工具(二)之解決方案2

發布時間:2020-09-07 00:26:49 來源:網絡 閱讀:590 作者:dbapower 欄目:關系型數據庫

解決方案二:在emp2empno列上面創建索引,再執行share_pool_sql_1.sh腳本,查看sp報告

8.1emp2empno列上創建索引

sys@TESTDB12>create index ind_empno on scott.emp2(empno);

8.2 重新執行share_pool_sql_1.sh腳本并重新開啟statspack自動快照

 

{oracle@Redhat55.cuug.net:/home/oracle/script/bin}$sh share_pool_sql_1.sh

SQL>@?/rdbms/admin/spauto

 

8.3生成statspack報告

perfstat@TESTDB12>selectsnap_id,snap_time,snap_level from stats$snapshot order by snap_time;

 

perfstat@TESTDB12>selectsnap_id,snap_time,snap_level from stats$snapshot order by snap_time;

 

   SNAP_ID SNAP_TIME SNAP_LEVEL

---------- -------------------

         1 28-JUL-14          7

        11 28-JUL-14          7

        21 28-JUL-14          7

        31 28-JUL-14          7

        41 29-JUL-14          7

        51 29-JUL-14          7

        61 29-JUL-14          7

        71 29-JUL-14          7

        81 29-JUL-14          7

        91 29-JUL-14          7

       101 29-JUL-14          7

       111 29-JUL-14          7

       121 29-JUL-14          7

       131 29-JUL-14          7

       141 29-JUL-14          7

       151 29-JUL-14          7

       161 29-JUL-14          7

       171 29-JUL-14          7

       181 29-JUL-14          7

       191 29-JUL-14          7

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Enter value for begin_snap: 131

Enter value for end_snap: 141

Enter value for report_name:

 

Specify the Begin and End Snapshot Ids

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Enter value for begin_snap: 141

Enter value for end_snap: 151

Enter value for report_name:

 

Specify the Begin and End Snapshot Ids

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Enter value for begin_snap: 151

Enter value for end_snap: 161

Enter value for report_name:

 

Specify the Begin and End Snapshot Ids

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Enter value for begin_snap: 161

Enter value for end_snap: 171

Enter value for report_name:

8.4通過新生成的4statspack報告對比各個時間段的數據緩沖區的命中率和庫緩沖區的命中率:

時間

Buffer  Hit%

Library Hit%

05:19:01~ 05:34:01

99.99

89.78

05:34:01 ~05:49:00  

99.99

89.72

05:49:00 ~ 06:04:05

99.98

89.45

06:04:05 ~06:13:00

99.95

88.79

 emp2empno列上創建索引后通過對比發現數據緩沖區的命中率明顯得到了改善,達到了的99%以上;而庫緩沖區的命中率也得到小幅度提升

8.5查看Top 5 Timed Events找出4 個報告中各個時間段跟磁盤I/O相關的等待事件

時間

name

Wait(s)

Time(s)

05:19:01~ 05:34:01

log file parallel  write

45,110

54


log file sync

6,240

46


os thread startup

34

5


control file  parallel write  

332

3

05:34:01 ~05:49:00

log file parallel  write

 48,413

36


log file sync

3,563

28


os thread startup

33

5


db file sequential  read

2,018

2

05:49:00 ~ 06:04:05

log file parallel  write

49,564

23


log file sync

455

15


db file sequential  read

3,955

9


os thread startup

39

6

06:04:05 ~06:13:00

log file parallel  write

 28,273

8


db  file sequential read 

 2,928

5


log  file sync

231

4


os  thread startup

21

3

通過4個報告的對比Top 5 Timed Eventsdirect path read不見了,說明解決了全表掃描等待I\O的問題;但log file parallel writelog file sync的磁盤I/O都還比較大,而且新增了control fileparallel write I/O,沒有什么大的耗資源的任務,說明系統性能得以提升

 

8.6造成物理讀最大的前幾個sql語句在報告中未找到,用sql語句查詢得出這些語句:select sql_text from v$sqlwhere disk_reads=(select max(disk_reads) from v$sql);

時間

 Executions

Rows  per Exec

Sql語句

05:19:01~ 05:34:01

10,840

16.1

select /*+ rule */  bucket, endpoint, col#, epvalue from histgrm$

 where obj#=:1 and intcol#=:2 and row#=:3  order by bucket









05:34:01 ~05:49:00

12,565

16.1

select /*+ rule */  bucket, endpoint, col#, epvalue from histgrm$

 where obj#=:1 and intcol#=:2 and row#=:3  order by bucket





05:49:00 ~ 06:04:05

15,112

 16.0

select /*+ rule */  bucket, endpoint, col#, epvalue from histgrm$

 where obj#=:1 and intcol#=:2 and row#=:3  order by bucket





06:04:05 ~06:13:00

 20,814

16.4

select /*+ rule */  bucket, endpoint, col#, epvalue from histgrm$

 where obj#=:1 and intcol#=:2 and row#=:3  order by bucket





通過對比各時間段最消耗資源的SQL語句,發現仍有相同或相似的執行計劃,應該使用綁定變量,來提高執行效率。

 

 

生成語句的執行計劃: set autotrace traceonly   select * from scott.emp2

idle>select *from scott.emp2 where empno=1484;

 

 

Execution Plan

----------------------------------------------------------

Plan hash value:2918945472

 

-----------------------------------------------------------------------------------------

| Id  | Operation                   | Name      | Rows | Bytes | Cost (%CPU)| Time     |

-----------------------------------------------------------------------------------------

|   0 | SELECT STATEMENT            |           |    1 |    48 |     4  (0)| 00:00:01 |

|   1 | TABLE ACCESS BY INDEX ROWID| EMP2     |     1 |    48 |    4   (0)| 00:00:01 |

|*  2 |  INDEX RANGE SCAN          |IND_EMPNO |     1 |       |    3   (0)| 00:00:01 |

-----------------------------------------------------------------------------------------

 

PredicateInformation (identified by operation id):

---------------------------------------------------

 

   2 - access("EMPNO"=1484)

 

 

Statistics

----------------------------------------------------------

         55 recursive calls

          0 db block gets

         78 consistent gets

          4 physical reads

          0 redo size

       1033 bytes sent via SQL*Net to client

        523 bytes received via SQL*Net from client

          2 SQL*Net roundtrips to/from client

          7 sorts (memory)

          0 sorts (disk)

          1 rows processed

8.7查看Buffer Pool Advisory并把Buffer cache的大小設置為推薦的大小

05:19:01~ 05:34:01時間段的Buffer PoolAdvisory

 

                                   Est

                                  Phys      Estimated                   Est

    Size for Size      Buffers   Read    Phys Reads     Est Phys % dbtime

P    Est (M) Factr  (thousands) Factr    (thousands)    Read Time for Rds

--- -------- ----------------- ------ -------------- ------------ --------

D          4   .1            0    8.0            261          345      5.2

D          8   .2            1    1.1             35           37       .6

D         12   .3            1    1.1             34           36       .5

D         16   .3            2    1.0             33           35       .5

D         20   .4            2    1.0             33           34       .5

D         24   .5            3    1.0             33           34       .5

D         28   .6            3    1.0             33           34       .5

D         32   .7            4    1.0             33           33       .5

D         36   .8            4    1.0             33           33       .5

D         40   .8            5    1.0             32           33       .5

D         44   .9            5    1.0             32           33       .5

D         48  1.0            6    1.0             32           33       .5

D         52  1.1            6    1.0             32           33       .5

D         56  1.2            7    1.0             32           33       .5

D         60  1.3            7    1.0             32           33       .5

D         64  1.3            8    1.0             32           33       .5

D         68  1.4            8    1.0             32           33       .5

D         72  1.5            9    1.0             32           33       .5

D         76  1.6            9    1.0             32           33       .5

D         80  1.7           10    1.0             32           33       .5

05:34:01 ~05:49:00時間段的Buffer PoolAdvisory

                                   Est

                                  Phys      Estimated                   Est

    Size for Size      Buffers   Read    Phys Reads     Est Phys % dbtime

P    Est (M) Factr  (thousands) Factr    (thousands)    Read Time for Rds

--- -------- ----------------- ------ -------------- ------------ --------

D          4   .1            0    7.8            273          357      5.1

D          8   .2            1    1.1             37           39       .6

D         12   .3            1    1.0             37           38       .5

D         16   .3            2    1.0             36           37       .5

D         20   .4            2    1.0             35           37       .5

D         24   .5            3    1.0             35           36       .5

D         28   .6            3    1.0             35           36       .5

D         32   .7            4    1.0             35           36       .5

D         36   .8            4    1.0             35           36       .5

D         40   .8            5    1.0             35           36       .5

D         44   .9            5    1.0             35           36       .5

D         48  1.0            6    1.0             35           36       .5

D         52  1.1            6    1.0             35           36       .5

D         56  1.2            7    1.0             35           36       .5

D         60  1.3            7    1.0             35           36       .5

D         64  1.3            8    1.0             35           36       .5

D         68  1.4            8    1.0             35           36       .5

D         72  1.5            9    1.0             35           36       .5

D         76  1.6            9    1.0             35           36       .5

D         80  1.7           10    1.0             35           36       .5

      

05:49:00 ~ 06:04:05時間段的Buffer PoolAdvisory

                                   Est

                                  Phys      Estimated                   Est

    Size for Size      Buffers   Read    Phys Reads     Est Phys % dbtime

P    Est (M) Factr  (thousands) Factr    (thousands)    Read Time for Rds

--- -------- ----------------- ------ -------------- ------------ --------

D          4   .1            0    7.6            302          438      6.0

D          8   .2            1    1.1             42           49       .7

D         12   .3            1    1.0             41           48       .7

D         16   .3            2    1.0             40           47       .6

D         20   .4            2    1.0             40           46       .6

D         24   .5            3    1.0             40           46       .6

D         28   .6            3    1.0             40           46       .6

D         32   .7            4    1.0             40           46       .6

D         36   .8            4    1.0             40           46       .6

D         40   .8            5    1.0             40           46       .6

D         44   .9            5    1.0             40           46       .6

D         48  1.0            6    1.0             40           46       .6

D         52  1.1            6    1.0             40           46       .6

D         56  1.2            7    1.0             40           46       .6

D         60  1.3            7    1.0             40           46       .6

D         64  1.3            8    1.0             40           46       .6

D         68  1.4            8    1.0             40           46       .6

D         72  1.5            9    1.0             40           46       .6

D         76  1.6            9    1.0             40           46       .6

D         80  1.7           10    1.0             40           46       .6

 

 

06:04:05 ~06:13:00時間段的Buffer PoolAdvisory

                                   Est

                                  Phys      Estimated                   Est

    Size for Size      Buffers   Read    Phys Reads     Est Phys % dbtime

P    Est (M) Factr  (thousands) Factr    (thousands)    Read Time for Rds

--- -------- ----------------- ------ -------------- ------------ --------

D          4   .1            0    7.6            338          497      6.6

D          8   .2            1    1.0             47           56       .7

D         12   .3            1    1.0             46           55       .7

D         16   .3            2    1.0             45           54       .7

D         20   .4            2    1.0             45           54       .7

D         24   .5            3    1.0             45           54       .7

D         28   .6            3    1.0             45           53       .7

D         32   .7            4    1.0             45           53       .7

D         36   .8            4    1.0             45           53       .7

D         40   .8            5    1.0             45           53       .7

D         44   .9            5    1.0             45           53       .7

D         48  1.0            6    1.0             45           53       .7

D         52  1.1            6    1.0             45           53       .7

D         56  1.2            7    1.0             45           53       .7

D         60  1.3            7    1.0             45           53       .7

D         64  1.3            8    1.0             45           53       .7

D         68  1.4            8    1.0             45           53       .7

D         72  1.5            9    1.0             45           53       .7

D         76  1.6            9    1.0             45           53       .7

D         80  1.7           10    1.0             45           53       .7

 

通過以上4個時間段中Buffer Pool Advisory建議可以看的出來,對于增加Buffer cache的大小對性能的影響并不明顯。

 

 

8.8查看Time Model System Stats

05:19:01~  05:34:01時間段Time Model System Stats

Statistic                                       Time  (s) % DB time

-----------------------------------  -------------------- ---------

DB CPU                                              440.5     119.9

parse time  elapsed                                  158.5      43.1

sql execute  elapsed time                            145.1      39.5

hard parse elapsed time                            135.0      36.8

connection  management call elapsed                  108.8      29.6

PL/SQL execution  elapsed time                         5.7       1.6

hard parse  (sharing criteria) elaps                   1.3        .3

hard parse (bind  mismatch) elapsed                    1.2        .3

PL/SQL compilation  elapsed time                       0.8        .2

repeated bind  elapsed time                            0.4        .1

sequence load  elapsed time                            0.1        .0

DB time                                             367.4

background elapsed  time                             75.1

background cpu  time                                  20.1

05:34:01  ~05:49:00時間段Time Model System Stats

Statistic                                       Time  (s) % DB time

-----------------------------------  -------------------- ---------

DB CPU                                              455.9     124.3

parse time  elapsed                                  155.5      42.4

sql execute  elapsed time                            149.9      40.9

hard parse elapsed time                            128.2      35.0

connection  management call elapsed                  104.6      28.5

PL/SQL execution  elapsed time                         6.8       1.9

hard parse  (sharing criteria) elaps                   2.5        .7

hard parse (bind  mismatch) elapsed                    2.4        .7

PL/SQL compilation  elapsed time                       0.8        .2

repeated bind  elapsed time                            0.5        .1

sequence load  elapsed time                            0.3        .1

DB time                                             366.8

background elapsed  time                             54.4

background cpu  time                                  20.1

05:49:00 ~  06:04:05時間段Time Model System Stats

Statistic                                       Time  (s) % DB time

-----------------------------------  -------------------- ---------

DB CPU                                              463.3     122.2

parse time  elapsed                                  160.9      42.4

sql execute  elapsed time                            158.6      41.9

hard parse elapsed time                            133.8      35.3

connection  management call elapsed                  103.6      27.3

PL/SQL execution  elapsed time                         7.3       1.9

hard parse  (sharing criteria) elaps                   2.1        .6

hard parse (bind  mismatch) elapsed                    1.9        .5

PL/SQL compilation  elapsed time                       1.1        .3

repeated bind  elapsed time                            0.5        .1

sequence load  elapsed time                            0.2        .0

DB time                                             379.0

background elapsed  time                             52.7

background cpu  time                                  23.0

06:04:05 ~06:13:00時間段Time Model  System Stats

Statistic                                       Time  (s) % DB time

-----------------------------------  -------------------- ---------

DB CPU                                              269.2     119.5

parse time  elapsed                                  105.7      46.9

sql execute  elapsed time                            102.9      45.6

hard parse elapsed time                             89.9      39.9

connection  management call elapsed                   58.2      25.8

PL/SQL execution  elapsed time                         4.0       1.8

hard parse  (sharing criteria) elaps                   2.0        .9

hard parse (bind  mismatch) elapsed                    1.6        .7

PL/SQL compilation  elapsed time                       1.1        .5

repeated bind  elapsed time                            0.6        .3

sequence load  elapsed time                            0.1        .1

DB time                                             225.4

background elapsed  time                             19.6

background cpu  time                                  12.2

通過對比4個報告各個時間段中的Time Model System Stats,發現產生的硬解析明顯增加了。

 

8.9查看Latch Sleep breakdown

05:19:01~  05:34:01時間段的Latch Sleep breakdown

Latch Name                        Requests       Misses      Sleeps        Gets

--------------------------  --------------- ------------ ----------- -----------

shared pool                      3,787,761            4           4           0

05:34:01  ~05:49:00時間段的Latch Sleep breakdown

Latch Name                        Requests       Misses      Sleeps        Gets

--------------------------  --------------- ------------ ----------- -----------

shared pool                      4,107,841            5           5           0

JS Sh mem  access                         3            1           1           0

enqueue hash  chains                320,877            1           1           0

05:49:00 ~  06:04:05時間段的Latch Sleep breakdown

Latch Name                        Requests       Misses      Sleeps        Gets

--------------------------  --------------- ------------ ----------- -----------

shared pool                      4,257,852            8           8           0

row cache  objects                3,956,966            3           3           0

call  allocation                     110,566            1           1           0

redo  allocation                      99,927            1           1           0

06:04:05 ~06:13:00時間段的Latch Sleep breakdown

Latch Name                        Requests       Misses      Sleeps        Gets

--------------------------  --------------- ------------ ----------- -----------

shared pool                      2,595,386            6           6           0

row cache  objects                2,500,734            1           1           0

通過以上4sp報告各個時間段的Latch Sleepbreakdown的內容,發現cache bufferslru chain已經沒有了,但是shared pool次數上來了。


向AI問一下細節

免責聲明:本站發布的內容(圖片、視頻和文字)以原創、轉載和分享為主,文章觀點不代表本網站立場,如果涉及侵權請聯系站長郵箱:is@yisu.com進行舉報,并提供相關證據,一經查實,將立刻刪除涉嫌侵權內容。

AI

重庆市| 叙永县| 上蔡县| 板桥市| 渝中区| 运城市| 平和县| 清镇市| 忻州市| 荆州市| 马尔康县| 策勒县| 湖州市| 乌鲁木齐县| 池州市| 陆良县| 宜兰市| 武安市| 乌拉特后旗| 石城县| 白山市| 安国市| 富平县| 仪陇县| 龙州县| 开化县| 新晃| 新密市| 弋阳县| 正蓝旗| 侯马市| 东兴市| 大城县| 三亚市| 泽普县| 都昌县| 乐都县| 阳高县| 双城市| 句容市| 张家界市|