分类: Oracle
2010-03-16 09:59:40
本文是关于Oracle数据库调试与优化方面的文章,主要介绍Oracle数据库中命中率相关的问题,包括不同的算法之间性能的比对。
关于Oracle中各个命中率的计算以及相关的调优
1)Library Cache的命中率:
.计算公式:Library Cache Hit Ratio = sum(pinhits) / sum(pins)
- SQL>SELECT SUM(pinhits)/sum(pins)
- FROM V$LIBRARYCACHE;
通常在98%以上,否则,需要要考虑加大共享池,绑定变量,修改cursor_sharing等参数。
2)计算共享池内存使用率:
- SQL>SELECT (1 - ROUND(BYTES / (&TSP_IN_M * 1024 * 1024), 2)) * 100 || '%'
- FROM V$SGASTAT
- WHERE NAME = 'free memory' AND POOL = 'shared pool';
其中: &TSP_IN_M是你的总的共享池的SIZE(M)
共享池内存使用率,应该稳定在75%-90%间,太小浪费内存,太大则内存不足。
查询空闲的共享池内存:
- SQL>SELECT * FROM V$SGASTAT
- WHERE NAME = 'free memory' AND POOL = 'shared pool';
3)db buffer cache命中率:
计算公式:Hit ratio = 1 - [physical reads/(block gets + consistent gets)]
- SQL>SELECT NAME, PHYSICAL_READS, DB_BLOCK_GETS, CONSISTENT_GETS,
- 1 - (PHYSICAL_READS / (DB_BLOCK_GETS + CONSISTENT_GETS)) "Hit Ratio"
- FROM V$BUFFER_POOL_STATISTICS
- WHERE NAME='DEFAULT';
通常应在90%以上,否则,需要调整,加大DB_CACHE_SIZE
另外一种计算命中率的方法(摘自ORACLE官方文档<<数据库性能优化>>):
命中率的计算公式为:
- Hit Ratio = 1 - ((physical reads - physical reads direct - physical reads direct (lob)) / (db block gets +
- consistent gets - physical reads direct - physical reads direct (lob))
分别代入上一查询中的结果值,就得出了Buffer cache的命中率
- SQL>SELECT NAME, VALUE
- FROM V$SYSSTAT
- WHERE NAME IN('session logical reads',
- 'physical reads',
- 'physical reads direct',
- 'physical reads direct (lob)',
- 'db block gets', 'consistent gets');
4)数据缓冲区命中率:
- SQL> select value from v$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';
这里命中率的计算应该是
令 x = physical reads direct + physical reads direct (lob)
命中率 =100 - ( physical reads - x) / (consistent gets + db block gets - x)*100
通常如果发现命中率低于90%,则应该调整应用可可以考虑是否增大数据缓冲区
5)共享池的命中率:
- SQL> select sum(pinhits-reloads)/sum(pins)*100 "hit radio" from v$librarycache;
假如共享池的命中率低于95%,就要考虑调整应用(通常是没使用bind var )或者增加内存
6)计算在内存中排序的比率:
- SQL>SELECT * FROM v$sysstat t WHERE NAME='sorts (memory)';—查询内存排序数
- SQL>SELECT * FROM v$sysstat t WHERE NAME='sorts (disk)';—查询磁盘排序数
- --caculate sort in memory ratio
- SQL>SELECT round(&sort_in_memory/(&sort_in_memory+&sort_in_disk),4)*100||'%' FROM dual;
此比率越大越好,太小整要考虑调整,加大PGA
7)PGA的命中率:
计算公式:BP x 100 / (BP + EBP)
BP: bytes processed
EBP: extra bytes read/written
- SQL>SELECT * FROM V$PGASTAT WHERE NAME='cache hit percentage';
或者从OEM的图形界面中查看
我们可以查看一个视图以获取Oracle的建议值:
- SQL>SELECT round(PGA_TARGET_FOR_ESTIMATE/1024/1024) target_mb,
- ESTD_PGA_CACHE_HIT_PERCENTAGE cache_hit_perc,
- ESTD_OVERALLOC_COUNT
- FROM V$PGA_TARGET_ADVICE;
- The output of this query might look like the following:
- TARGET_MB CACHE_HIT_PERC ESTD_OVERALLOC_COUNT
- ---------- -------------- --------------------
- 63 23 367
- 125 24 30
- 250 30 3
- 375 39 0
- 500 58 0
- 600 59 0
- 700 59 0
- 800 60 0
- 900 60 0
在此例中:PGA至少要分配375M
我个人认为PGA命中率不应该低于50%
以下的SQL统计sql语句执行在三种模式的次数: optimal memory size, one-pass memory size, multi-pass memory size:
- SQL>SELECT name profile, cnt, decode(total, 0, 0, round(cnt*100/total,4)) percentage
- FROM (SELECT name, value cnt, (sum(value) over ()) total FROM V$SYSSTAT WHERE name like 'workarea exec%');
8)共享区字典缓存区命中率
计算公式:SUM(gets - getmisses - usage -fixed) / SUM(gets)
命中率应大于0.85
- SQL>select sum(gets-getmisses-usage-fixed)/sum(gets)
- from v$rowcache;
9)数据高速缓存区命中率
计算公式:1-(physical reads / (db block gets + consistent gets))
命中率应大于0.90最好
- SQL>select name,value
- from v$sysstat
- where name in ('physical reads','db block gets','consistent gets');
10)共享区库缓存区命中率
计算公式:SUM(pins - reloads) / SUM(pins)
命中率应大于0.99
- SQL>select sum(pins-reloads)/sum(pins)
- from v$librarycache;
11)检测回滚段的争用
SUM(waits)值应小于SUM(gets)值的1%
- SQL>select sum(gets),sum(waits),sum(waits)/sum(gets)
- from v$rollstat;
12)检测回滚段收缩次数
- SQL>select name,shrinks
- from v$rollstat, v$rollname
- where v$rollstat.usn = v$rollname.usn;
- -----------------------------------------------------------------------------
几个常用的检查语句
1. 查找排序最多的SQL:
- SQL>SELECT HASH_VALUE, SQL_TEXT, SORTS, EXECUTIONS
- FROM V$SQLAREA
- ORDER BY SORTS DESC;
2.查找磁盘读写最多的SQL:
- SQL>SELECT * FROM
- (SELECT sql_text,disk_reads "total disk" , executions "total exec",disk_reads/executions "disk/exec" FROM v$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(实际上也是按磁盘读写来排序的):
- SQL>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_$sql
- where command_type != 47)
- where disk_reads > 50 * executions) s,sys.v_$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:
- SQL>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$sqlarea
- Where executions>0 and buffer_gets >0 and (buffer_gets-disk_reads)/buffer_gets<0.8
- Order by 4 desc;
5、根据sid查看对应连接正在运行的sql
- SQL>select /*+ push_subq */command_type,sql_text,sharable_mem,persistent_mem,
- runtime_mem,sorts,version_count,loaded_versions,open_versions,users_opening,executions,
- users_executing,loads,first_load_time,invalidations,parse_calls,disk_reads,buffer_gets,
- rows_processed,sysdate start_time,sysdate finish_time,’>’||address sql_address,’N’status
- From v$sqlarea
- Where address=(select sql_address from v$session where sid=&sid);
- ***************Oracle 缓冲区命中率低的分析及解决办法******************
首先确定下面的查询结果:
1,缓冲区命中率的查询(是否低于90%):
- select round((1 - sum(decode(name,'physical reads',value,0)) /
- (sum(decode(name,'db block gets',value,0)) + sum(decode(name,'consistent gets',value,0))) ),4) *100 || '%' chitrati
- from v$sysstat;
2,使用率的查询(有无free状态的数据快.):
- select count(*), status from v$bh group by status ;
3,相关等待事件的查询(是否有相关等待事件)
- select event,total_waits from v$system_event where event in ('free buffer waits');
4,当前大小(是否已经很大)
- select value/1024/1024 cache_size from v$parameter where name='db_cache_size'
5,top等待事件分析(Db file scatered read的比率是否大)
- select event ,total_waits,suml
- from
- (select event,total_waits,round(total_waits/sumt*100,2)||'%' suml
- from
- (select event,total_waits from v$system_event ),
- (select sum(total_waits) sumt from v$system_event)
- order by total_waits desc)
- where rownum<6
- and event not like 'rdbms%'
- and event not like 'pmon%'
- and event not like 'SQL*Net%'
- and event not like 'smon%';
6,db_cache_advice建议值(9i后的新特性,可以根据他更好的调整cache_size)
- select block_size,size_for_estimate,size_factor,estd_physical_reads from v$db_cache_advice;
说明分析:
缓冲区命中率(低于90的命中率就算比较低的).
没有free不一定说明需要增加,还要结合当前cache_size的大小(我们是否还可以再增大,是否有需要增加硬件,增加开销),
空闲缓冲区等待说明进程找不到空闲缓冲区,并通过写出灰缓冲区,来加速数据库写入器生成空闲缓冲区,当DBWn将块写入磁盘后,灰数据缓冲区将被释放,以便重新使用.产生这种原因主要是:
1,DBWn可能跟不上写入灰缓冲区:i/0系统较慢,尽量将文件均匀的分布于所有设备,
2,缓冲区过小或过大。
3,可以增加db_writer_processes数量。
4,可能有很大的一个事物,或者连续的大事物
我们需要长期观察这个事件是否长期存在并数值一直在增大,如果一直在增大,则说明需要增大db_cache大小.或优化sql.
数据分散读等待,通常表现存在着与全表扫描相关的等待,逻辑读时,在内存中进行的全表扫描一般是零散地,而并非连续的被分散到缓冲区的各个部分,可能有索引丢失,或被仰制索引的存在。该等待时间在数据库会话等待多块io读取结束的时候产生,并把指定的块数离散的分布在数据缓冲区。这意味这全表扫描过多,或者io不足或争用,
存在这个事件,多数都是问题的,这说明大量的全部扫描而未采用索引.
db_cache_advice对我们调整db_cache_size大小有一定的帮助,但这只是一个参考,不一定很精确。
通过上面6种情况的综合分析,判断是否需要增加大cache_size. 或者把常用的(小)表放到keep区。
但多数的时候做这些不会解决质的问题,
而真正的问题主要是对sql语句的优化(如:是否存在大量的全表扫描等)
索引是在不需要改变程序的情况下,对数据库性能,sql语句提高的最实用的方法.
我在生产中遇到过类似的问题,200M的cache_size,命中率很低21%,但通过对sql语句的优化(添加索引,避免全表扫描),命中率增加到96%,程序运行时间由原来的2小时减少到不到10分钟.
这就提到了怎么定位高消耗的sql问题.全表扫描的问题,在这里不做细致的解说,这里只说明方法,我会在相关的章节专门介绍怎么使用这些工具
1,sql_trace跟踪session.用tkprof 分别输出磁盘读,逻辑读,运行时间长的sql进行优化.这些高消耗的sql一般都伴随着全表扫描.
2,statspack分析.在系统繁忙时期进行时间点的统计分析,产看TOP事件是否有Db file scatered read.并查看TOP sql语句是否存在问题等.
注:电脑学习网首发。
还要说一句:当然在硬件允许的情况下,尽量增大db_cache_size 减少磁盘读,但并不是越大越好,一定要根据自己的库数据量的程度来调节,因为大的db_cache_size同样会增大数据库管理的开销,当然可能开销并不会明显的影响数据库的性能,硬件价格也越来越低,这就需要我们具体问题具体分析了,在我看来物尽其用就最好了,尽量不要浪费,找到问题的本质。调优是一件很艺术的事。
***********************Oracle数据库缓冲区命中率*****************
1、查看Oracle数据库缓冲区命中率
- 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$sysstat c
- where a.statistic# = 40 and b.statistic# = 41
- and c.statistic# = 42;
2、Tags: oracle
数据库缓冲区命中率:
- sql>select value from v$sysstat where name ='physical reads';
- value
- 3714179
- sql>select value from v$sysstat where name ='physical reads direct';
- value
- 0
- sql>select value from v$sysstat where name ='physical reads direct(lob)';
- value
- 0
- sql>select value from v$sysstat where name ='consistent gets';
- value
- 856309623
- sql>select value from v$sysstat where name ='db block gets';
- value
- 19847790
这里命中率的计算应该是
令x=physical reads direct + physical reads direct(lob)
命中率=100-(physical reads -x)/(consistent gets +db block gets -x)*100
通常如果发现命中率低于90%,则应该调整应用可以考虑是否增大数据加
共享池的命中率
- sql> select sum(pinhits)/sum(pins)*100 "hit radio" from v$librarycache;
如果共享池的命中率低于95%就要考虑调整应用(通常是没应用bind var)或者增加内存。
关于排序部分
- sql> select name,value from v$sysstat where name like '%sort%';
如果我们发现sorts(disk)/(sorts(memory)+sorts(disk))的比例过高,则通常意味着sort_area_size部分内存教较小,可考虑调整相应的参数。
关于log_buffer
- sql>select name,value from v$sysstat where name in ('redo entries','redo buffer allocation retries');
假如redo buffer allocation retries/redo entries的比例超过1%我们就可以考虑增加log_buffer.