本文是關(guān)于Oracle數(shù)據(jù)庫調(diào)試與優(yōu)化方面的文章，主要介紹Oracle數(shù)據(jù)庫中命中率相關(guān)的問題，包括不同的算法之間性能的比對(duì)。關(guān)于Oracle中各個(gè)命中率的計(jì)算以及相關(guān)的調(diào)優(yōu) 1) Library Cache的命中率: 計(jì)算公式:Library Cache Hit Ratio = sum(pinhits) / sum(pi
本文是關(guān)于Oracle數(shù)據(jù)庫調(diào)試與優(yōu)化方面的文章，主要介紹Oracle數(shù)據(jù)庫中命中率相關(guān)的問題，包括不同的算法之間性能的比對(duì)。關(guān)于Oracle中各個(gè)命中率的計(jì)算以及相關(guān)的調(diào)優(yōu)
1)Library Cache的命中率:計(jì)算公式:Library Cache Hit Ratio = sum(pinhits) / sum(pins)
|
1 |
SELECT SUM(pinhits)/sum(pins) FROM V$LIBRARYCACHE;
|
通常在98%以上，否則，需要要考慮加大共享池，綁定變量，修改cursor_sharing等參數(shù)。
2)計(jì)算共享池內(nèi)存使用率:
|
1
2 |
SELECT (1 - ROUND(BYTES / (&TSP_IN_M * 1024 * 1024), 2)) * 100 || '%'
FROM V$SGASTAT WHERE NAME = 'free memory' AND POOL = 'shared pool';
|
其中: &TSP_IN_M是你的總的共享池的SIZE(M)
共享池內(nèi)存使用率，應(yīng)該穩(wěn)定在75%-90%間，太小浪費(fèi)內(nèi)存，太大則內(nèi)存不足。
查詢空閑的共享池內(nèi)存:
|
1
2 |
SELECT * FROM V$SGASTAT WHERE
NAME = 'free memory' AND POOL = 'shared pool';
|
3)db buffer cache命中率:計(jì)算公式:Hit ratio = 1 - [physical reads/(block gets + consistent gets)]
|
1
2
3 |
SELECT NAME, PHYSICAL_READS, DB_BLOCK_GETS, CONSISTENT_GETS,
1 - (PHYSICAL_READS / (DB_BLOCK_GETS + CONSISTENT_GETS))
"Hit Ratio" FROM V$BUFFER_POOL_STATISTICS WHERE NAME='DEFAULT';
|
通常應(yīng)在90%以上，否則，需要調(diào)整,加大DB_CACHE_SIZE
外一種計(jì)算命中率的方法(摘自O(shè)RACLE官方文檔<<數(shù)據(jù)庫性能優(yōu)化>>):
命中率的計(jì)算公式為:
Hit Ratio = 1 - ((physical reads - physical reads direct - physical reads direct (lob)) / (db block gets + consistent gets - physical reads direct - physical reads direct (lob))
分別代入上一查詢中的結(jié)果值,就得出了Buffer cache的命中率
|
1
2
3
4 |
SELECT NAME, VALUE FROM V$SYSSTAT WHERE NAME IN(
'session logical reads', 'physical reads',
'physical reads direct', 'physical reads direct (lob)',
'db block gets', 'consistent gets');
|
4)數(shù)據(jù)緩沖區(qū)命中率：
|
1
2
3
4
5 |
SQL> select value from v$sysstat where name ='physical reads';
SQL> select value from v$sysstat where name ='physical reads direct';
SQL> select value from v$sysstat where name ='physical reads direct (lob)';
SQL> select value from v$sysstat where name ='consistent gets';
SQL> select value from v$sysstat where name = 'db block gets';
|
這里命中率的計(jì)算應(yīng)該是令 x = physical reads direct + physical reads direct (lob),命中率 =100 - ( physical reads - x) / (consistent gets + db block gets - x)*100,通常如果發(fā)現(xiàn)命中率低于90%,則應(yīng)該調(diào)整應(yīng)用可可以考慮是否增大數(shù)據(jù)緩沖區(qū)
5)共享池的命中率：
|
1
2 |
select sum(pinhits-reloads)/sum(pins)*100 "hit radio"
from v$librarycache;
|
假如共享池的命中率低于95%,就要考慮調(diào)整應(yīng)用(通常是沒使用bind var )或者增加內(nèi)存
6)計(jì)算在內(nèi)存中排序的比率:
|
1 |
SELECT * FROM v$sysstat t WHERE NAME='sorts (memory)';
|
查詢內(nèi)存排序數(shù)
|
1 |
SELECT * FROM v$sysstat t WHERE NAME='sorts (disk)';
|
查詢磁盤排序數(shù)caculate sort in memory ratio
|
1 |
SELECT round(&sort_in_memory/(&sort_in_memory+&sort_in_disk),4)*100||'%' FROM dual;
|
此比率越大越好,太小整要考慮調(diào)整,加大PGA
7)PGA的命中率:
計(jì)算公式:BP x 100 / (BP + EBP)
BP: bytes processed
EBP: extra bytes read/written
|
1 |
SELECT * FROM V$PGASTAT WHERE NAME='cache hit percentage';
|
或者從OEM的圖形界面中查看
我們可以查看一個(gè)視圖以獲取Oracle的建議值:
|
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15 |
SELECT round(PGA_TARGET_FOR_ESTIMATE/1024/1024) target_mb,
ESTD_PGA_CACHE_HIT_PERCENTAGE cache_hit_perc, ESTD_OVERALLOC_COUNT
FROM V$PGA_TARGET_ADVICE;
The output of this query might look like the following:
TARGET_MB CACHE_HIT_PERC ESTD_OVERALLOC_COUNT
63 23 367
125 24 30
250 30 3
375 39 0
500 58 0
600 59 0
700 59 0
800 60 0
900 60 0
|
在此例中:PGA至少要分配375M
我個(gè)人認(rèn)為PGA命中率不應(yīng)該低于50%
以下的SQL統(tǒng)計(jì)sql語句執(zhí)行在三種模式的次數(shù): optimal memory size, one-pass memory size, multi-pass memory size:
|
1
2
3 |
SELECT name profile, cnt, decode(total, 0, 0, round(cnt*100/total,4))
percentage FROM (SELECT name, value cnt, (sum(value) over ())
total FROM V$SYSSTAT WHERE name like 'workarea exec%');
|
8)共享區(qū)字典緩存區(qū)命中率
計(jì)算公式：SUM(gets - getmisses - usage -fixed) / SUM(gets)
命中率應(yīng)大于0.85
|
1 |
select sum(gets-getmisses-usage-fixed)/sum(gets) from v$rowcache;
|
9)數(shù)據(jù)高速緩存區(qū)命中率
計(jì)算公式：1-(physical reads / (db block gets + consistent gets))
命中率應(yīng)大于0.90最好
|
1 |
select name,value from v$sysstat where name in ('physical reads','db block gets','consistent gets');
|
10)共享區(qū)庫緩存區(qū)命中率
計(jì)算公式：SUM(pins - reloads) / SUM(pins)
命中率應(yīng)大于0.99
|
1 |
select sum(pins-reloads)/sum(pins) from v$librarycache;
|
11)檢測回滾段的爭用
SUM(waits)值應(yīng)小于SUM(gets)值的1%
|
1 |
select sum(gets),sum(waits),sum(waits)/sum(gets) from v$rollstat;
|
12)檢測回滾段收縮次數(shù)
|
1
2 |
select name,shrinks from v$rollstat, v$rollname
where v$rollstat.usn = v$rollname.usn;
|
幾個(gè)常用的檢查語句
1. 查找排序最多的SQL:
|
1
2 |
SELECT HASH_VALUE, SQL_TEXT, SORTS, EXECUTIONS
FROM V$SQLAREA ORDER BY SORTS DESC;
|
2.查找磁盤讀寫最多的SQL:
|
1
2
3
4 |
SELECT * FROM (SELECT sql_text,disk_reads "total disk" ,
executions "total exec",disk_reads/executions "disk/exec"
FROM v$sql WHERE executions>0 and is_obsolete='N' ORDER BY
4 desc) WHERE ROWNUM<11 ;
|
3.查找工作量最大的SQL(實(shí)際上也是按磁盤讀寫來排序的):
|
1
2
3
4
5
6
7
8 |
select substr(to_char(s.pct, '99.00'), 2) || '%' load,
s.executions executes,p.sql_text from(select address,disk_reads,executions,pct,rank() over
(order by disk_reads desc) ranking from (select
address,disk_reads,executions,100 * ratio_to_report
(disk_reads) over () pct from sys.v_$sql where
command_type != 47) where disk_reads > 50 * executions) s,
ys.v_$sqltext p where s.ranking <= 5 and p.address = s.address
order by 1, s.address, p.piece;
|
4. 用下列SQL工具找出低效SQL:
|
1
2
3
4
5 |
select executions,disk_reads,buffer_gets,round((buffer_gets-
disk_reads)/buffer_gets,2) Hit_radio,round(disk_reads/executions,2)
reads_per_run,sql_text From v$sqlarea Where executions>0 and
buffer_gets >0 and (buffer_gets-disk_reads)/buffer_gets<0.8
Order by 4 desc;
|
5、根據(jù)sid查看對(duì)應(yīng)連接正在運(yùn)行的sql
|
1
2
3
4
5
6
7 |
select /*+ push_subq */command_type,sql_text,sharable_mem,persistent_mem,
runtime_mem,sorts,version_count,loaded_versions,open_versions,
users_opening,executions,users_executing,loads,first_load_time,
invalidations,parse_calls,disk_reads,buffer_gets,rows_processed,
sysdate start_time,sysdate finish_time,’>’||address
sql_address,’N’status From v$sqlarea Where address=
(select sql_address from v$session where sid=&sid);
|
***************Oracle 緩沖區(qū)命中率低的分析及解決辦法******************
首先確定下面的查詢結(jié)果:
1,緩沖區(qū)命中率的查詢(是否低于90%):
|
1
2
3 |
select round((1 - sum(decode(name,'physical reads',value,0)) /
(sum(decode(name,'db block gets',value,0)) + sum(decode(name,'
consistent gets',value,0))) ),4) *100 || '%' chitrati from v$sysstat;
|
2,使用率的查詢(有無free狀態(tài)的數(shù)據(jù)快.):
|
1 |
select count(*), status from v$bh group by status ;
|
3,相關(guān)等待事件的查詢(是否有相關(guān)等待事件)
|
1
2 |
select event,total_waits from v$system_event where event
in ('free buffer waits');
|
4,當(dāng)前大小(是否已經(jīng)很大)
|
1
2 |
select value/1024/1024 cache_size from v$parameter
where name='db_cache_size'
|
5,top等待事件分析(Db file scatered read的比率是否大)
|
1
2
3
4
5
6
7 |
select event ,total_waits,suml from (select
event,total_waits,round(total_waits/sumt*100,2)||'%' suml
from (select event,total_waits from v$system_event ),
(select sum(total_waits) sumt from v$system_event)
order by total_waits desc) where rownum<6 and event
not like 'rdbms%' and event not like 'pmon%' and event
not like 'SQL*Net%' and event not like 'smon%';
|
6,db_cache_advice建議值(9i后的新特性,可以根據(jù)他更好的調(diào)整cache_size)
|
1
2 |
select block_size,size_for_estimate,size_factor,
estd_physical_reads from v$db_cache_advice;
|
說明分析:
緩沖區(qū)命中率(低于90的命中率就算比較低的).
沒有free不一定說明需要增加,還要結(jié)合當(dāng)前cache_size的大小(我們是否還可以再增大,是否有需要增加硬件,增加開銷),
空閑緩沖區(qū)等待說明進(jìn)程找不到空閑緩沖區(qū)，并通過寫出灰緩沖區(qū)，來加速數(shù)據(jù)庫寫入器生成空閑緩沖區(qū)，當(dāng)DBWn將塊寫入磁盤后，灰數(shù)據(jù)緩沖區(qū)將被釋放，以便重新使用.產(chǎn)生這種原因主要是:
1，DBWn可能跟不上寫入灰緩沖區(qū)：i/0系統(tǒng)較慢，盡量將文件均勻的分布于所有設(shè)備，
2，緩沖區(qū)過小或過大。
3，可以增加db_writer_processes數(shù)量。
4,可能有很大的一個(gè)事物，或者連續(xù)的大事物
我們需要長期觀察這個(gè)事件是否長期存在并數(shù)值一直在增大,如果一直在增大,則說明需要增大db_cache大小.或優(yōu)化sql.
數(shù)據(jù)分散讀等待,通常表現(xiàn)存在著與全表掃描相關(guān)的等待，邏輯讀時(shí)，在內(nèi)存中進(jìn)行的全表掃描一般是零散地，而并非連續(xù)的被分散到緩沖區(qū)的各個(gè)部分，可能有索引丟失，或被仰制索引的存在。該等待時(shí)間在數(shù)據(jù)庫會(huì)話等待多塊io讀取結(jié)束的時(shí)候產(chǎn)生，并把指定的塊數(shù)離散的分布在數(shù)據(jù)緩沖區(qū)。這意味這全表掃描過多，或者io不足或爭用，
存在這個(gè)事件,多數(shù)都是問題的,這說明大量的全部掃描而未采用索引.
db_cache_advice對(duì)我們調(diào)整db_cache_size大小有一定的幫助,但這只是一個(gè)參考，不一定很精確。
通過上面6種情況的綜合分析,判斷是否需要增加大cache_size. 或者把常用的(小)表放到keep區(qū)。
但多數(shù)的時(shí)候做這些不會(huì)解決質(zhì)的問題,
而真正的問題主要是對(duì)sql語句的優(yōu)化(如:是否存在大量的全表掃描等)
索引是在不需要改變程序的情況下,對(duì)數(shù)據(jù)庫性能，sql語句提高的最實(shí)用的方法.
我在生產(chǎn)中遇到過類似的問題,200M的cache_size,命中率很低21%,但通過對(duì)sql語句的優(yōu)化(添加索引,避免全表掃描),命中率增加到96%,程序運(yùn)行時(shí)間由原來的2小時(shí)減少到不到10分鐘.
這就提到了怎么定位高消耗的sql問題.全表掃描的問題,在這里不做細(xì)致的解說,這里只說明方法,我會(huì)在相關(guān)的章節(jié)專門介紹怎么使用這些工具
1,sql_trace跟蹤session.用tkprof 分別輸出磁盤讀,邏輯讀,運(yùn)行時(shí)間長的sql進(jìn)行優(yōu)化.這些高消耗的sql一般都伴隨著全表掃描.
2,statspack分析.在系統(tǒng)繁忙時(shí)期進(jìn)行時(shí)間點(diǎn)的統(tǒng)計(jì)分析,產(chǎn)看TOP事件是否有Db file scatered read.并查看TOP sql語句是否存在問題等.
還要說一句：當(dāng)然在硬件允許的情況下，盡量增大db_cache_size 減少磁盤讀，但并不是越大越好，一定要根據(jù)自己的庫數(shù)據(jù)量的程度來調(diào)節(jié)，因?yàn)榇蟮膁b_cache_size同樣會(huì)增大數(shù)據(jù)庫管理的開銷，當(dāng)然可能開銷并不會(huì)明顯的影響數(shù)據(jù)庫的性能，硬件價(jià)格也越來越低，這就需要我們具體問題具體分析了，在我看來物盡其用就最好了，盡量不要浪費(fèi)，找到問題的本質(zhì)。調(diào)優(yōu)是一件很藝術(shù)的事。
***********************Oracle數(shù)據(jù)庫緩沖區(qū)命中率*****************
1、查看Oracle數(shù)據(jù)庫緩沖區(qū)命中率
|
1
2
3
4 |
select a.value + b.value "logical_reads", c.value "phys_reads",
round(100 * ((a.value+b.value)-c.value) / (a.value+b.value))
"BUFFER HIT RATIO" from v$sysstat a, v$sysstat b, v$sysstat c
where a.statistic# = 40 and b.statistic# = 41 and c.statistic# = 42;
|
2、Tags: oracle
數(shù)據(jù)庫緩沖區(qū)命中率：
|
1
2
3
4
5
6
7
8
9
10
11
12
13
14 |
select value from v$sysstat where name ='physical reads';
value 3714179
select value from v$sysstat where name ='physical reads direct';
value 0
select value from v$sysstat where name ='physical reads direct(lob)';
value 0
select value from v$sysstat where name ='consistent gets';
value 856309623
select value from v$sysstat where name ='db block gets';
value 19847790
|
這里命中率的計(jì)算應(yīng)該是
令x=physical reads direct + physical reads direct(lob)
命中率=100-(physical reads -x)/(consistent gets +db block gets -x)*100
通常如果發(fā)現(xiàn)命中率低于90%，則應(yīng)該調(diào)整應(yīng)用可以考慮是否增大數(shù)據(jù)加
共享池的命中率
|
1 |
select sum(pinhits)/sum(pins)*100 "hit radio" from v$librarycache;
|
如果共享池的命中率低于95%就要考慮調(diào)整應(yīng)用(通常是沒應(yīng)用bind var)或者增加內(nèi)存。
關(guān)于排序部分
|
1 |
select name,value from v$sysstat where name like '%sort%';
|
如果我們發(fā)現(xiàn)sorts(disk)/(sorts(memory)+sorts(disk))的比例過高，則通常意味著sort_area_size部分內(nèi)存教較小，可考慮調(diào)整相應(yīng)的參數(shù)。
關(guān)于log_buffer
|
1
2 |
select name,value from v$sysstat where name
in ('redo entries','redo buffer allocation retries');
|
假如redo buffer allocation retries/redo entries的比例超過1%我們就可以考慮增加log_buffer.