调优5（SGA其他缓存区调整）

标签：class   java   c   int   get   com   

第五章 SGA其他缓存区调整

一、redo log buffer

1、redo log 的功能
1）Sever 进程在buffer cache修改数据块后，Oracle提倡‘先记后写’，对修改的数据块的改变生成log entries（日志条目),将日志条目按顺序写入log buffer；而对于脏块
先链接到检查点队列，等待dbwr 进程写入到datafile
2） 进程会及时的将其按顺序写入到redo log files。（日志条目的数据块大小一般和操作系统数据块大小一致，在Unix 和linux 下一般为512Bytes）
3）lgwr 在以下条件会写log entries：
A)、commit
B)、log buffer 空间1/3
C)、日志到1M时
D)、每隔3S
E)、在dbwr 写脏块前
--redo 日志及时的写入到redo log files，可以在instance 或 media recover 时，通过redo log 来对数据库进行recovery

2、redo log buffer 大小

09:59:26 SYS@ test1>select name ,bytes/1024/1024 from v$sgainfo;

NAME BYTES/1024/1024
-------------------------------- ---------------
Fixed SGA Size 1.16233444
Redo Buffers 2.8359375
----由于redo log buffer的空间循环使用，可以不用设太大的尺寸，一般默认值即可。

3、监控log buffer

10:07:36 SYS@ test1>select sid,event,seconds_in_wait,state
10:07:59 2 from v$session_wait
10:08:05 3 where event=‘log buffer space‘;

no rows selected
------‘log buffer space’ 事件，指的是当server 进程写日志条目到log buffer 时，没有可用的空间，而导致等待

10:10:58 SCOTT@ test1>begin
10:11:28 2 for i in 1..1000000
10:11:38 3 loop
10:11:41 4 execute immediate ‘insert into t1 values (:num)‘ using i;
10:11:46 5 end loop;
10:11:50 6 end;
10:11:51 7 /

PL/SQL procedure successfully completed.

10:13:10 SYS@ test1>r
1 select sid,event,seconds_in_wait,state
2 from v$session_wait
3* where event=‘log buffer space‘;
10:12:04 SYS@ test1>/

SID EVENT SECONDS_IN_WAIT STATE
---------- ---------------------------------------------------------------- --------------- -------------------
20 log buffer space 9 WAITED KNOWN TIME

Elapsed: 00:00:00.01
10:12:24 SYS@ test1>/

SID EVENT SECONDS_IN_WAIT STATE
---------- ---------------------------------------------------------------- --------------- -------------------
20 log buffer space 22 WAITED KNOWN TIME

Elapsed: 00:00:00.00
-----事务量增加，导致了“log buffer space” 的等待事件

10:17:55 SYS@ test1>select username,sid from v$session where username is not null;

USERNAME SID
------------------------------ ----------
SYS 39
SCOTT 28

Elapsed: 00:00:00.02
10:18:07 SYS@ test1>select SID,EVENT,TOTAL_WAITS WAIT_CLASS from v$session_event
10:18:29 2 where sid in(39,28) and event in (‘log file sync‘,‘log file parallel write‘);

SID EVENT WAIT_CLASS
---------- ---------------------------------------------------------------- ----------
39 log file sync 5

Elapsed: 00:00:00.03
10:18:30 SYS@ test1>

log file sync等待事件
当用户发出提交或回滚语句时会触发LGWR将重做记录写入联机日志文件，这种触发LGWR的方式叫做同步写（sync writes）触发，
而其他剩下的触发LGWR的方式叫做后台写（background writes）。log file sync等待事件只与sync writes有关，
而log file parallel write等待事件只与background writes有关。

-----调整（增加日志组或日志组的大小，或者增加log_buffer 的大小）

10:35:42 SYS@ test1>alter system set log_buffer =10240000 scope=spfile;

System altered.

10:38:53 SCOTT@ test1>
begin
10:39:03 SCOTT@ test1>for i in 1..1000000
loop
10:39:03 3 execute immediate ‘insert into t1 values (:num)‘ using i;
10:39:03 5 end loop;
10:39:03 6 end;
10:39:03 7 /

PL/SQL procedure successfully completed.

Elapsed: 00:01:39.52
10:40:44 SCOTT@ test1>

10:38:17 SYS@ test1>select sid,event,seconds_in_wait,state
10:39:14 2 from v$session_wait
10:39:14 3 where event=‘log buffer space‘;

no rows selected

Elapsed: 00:00:00.11
10:39:15 SYS@ test1>/

no rows selected

Elapsed: 00:00:00.04
10:39:42 SYS@ test1>/

no rows selected

Elapsed: 00:00:00.05
10:39:51 SYS@ test1>/

no rows selected

Elapsed: 00:00:00.09

10:43:18 SYS@ test1>select SID,EVENT,TOTAL_WAITS WAIT_CLASS from v$session_event
10:43:19 2 where sid in(39,22) and event in (‘log file sync‘,‘log file parallel write‘);

no rows selected

-------调整了redo buffer 空间后，log buffer space 等待事件被消除

-------查看日志条目的信息

10:20:30 SYS@ test1>select name,value from v$sysstat
10:20:51 2 where name in (‘redo buffer allocation retries‘,‘redo entries‘);

NAME VALUE
---------------------------------------------------------------- ----------
redo entries 1125660
redo buffer allocation retries 9

Elapsed: 00:00:00.01

------redo entries 生成的redo 日志条目的个数
------redo buffer allocation retries server 进程在写入日志条目时，没有足够的空间，重试的次数

--一般要求，重试次数应该为0，两者相比的比率小于1%

4、减少redo 的生成

1)、在做sql load 时采用direct 导入和nologging。
2)、在执行sql 语句时，使用nologging（但dml 操作无效）

------dml

0:32:36 SCOTT@ test1>update emp1 nologging set sal=sal+2000 where deptno=10;

12 rows updated.

Elapsed: 00:00:00.01

Execution Plan
----------------------------------------------------------
Plan hash value: 2879231406

-----------------------------------
| Id | Operation | Name |
-----------------------------------
| 0 | UPDATE STATEMENT | |
| 1 | UPDATE | EMP1 |
|* 2 | TABLE ACCESS FULL| EMP1 |
-----------------------------------

Predicate Information (identified by operation id):
---------------------------------------------------

2 - filter("DEPTNO"=10)

Note
-----
- rule based optimizer used (consider using cbo)

Statistics
----------------------------------------------------------
1 recursive calls
6 db block gets
23 consistent gets
0 physical reads
3036 redo size
681 bytes sent via SQL*Net to client
589 bytes received via SQL*Net from client
4 SQL*Net roundtrips to/from client
1 sorts (memory)
0 sorts (disk)
12 rows processed

10:32:44 SCOTT@ test1>

------在nologging 下仍有日志生成

------通过视图监控

10:36:58 SCOTT@ test1>select a.name,b.value from v$statname a,v$mystat b
10:37:05 2 where a.STATISTIC#=b.STATISTIC# and a.name like ‘%redo size%‘;

NAME VALUE
---------------------------------------------------------------- ----------
redo size 269785556

10:37:11 SCOTT@ test1>update emp1 nologging set sal=sal+2000 where deptno=10;

12 rows updated.

10:37:31 SCOTT@ test1>select a.name,b.value from v$statname a,v$mystat b
10:37:45 2 where a.STATISTIC#=b.STATISTIC# and a.name like ‘%redo size%‘;

NAME VALUE
---------------------------------------------------------------- ----------
redo size 269790652

------DDL

-------使用nologging 创建索引

10:40:32 SCOTT@ test1>create index t1_id_ind on t1(id) nologging;

Index created.

Elapsed: 00:00:04.53
10:41:12 SCOTT@ test1>select a.name,b.value from v$statname a,v$mystat b
10:41:25 2 where a.STATISTIC#=b.STATISTIC# and a.name like ‘%redo size%‘;

NAME VALUE
---------------------------------------------------------------- ----------
redo size 269910748

Elapsed: 00:00:00.00
10:41:33 SCOTT@ test1>

------正常创建索引

10:20:30 SYS@ test1>select name,value from v$sysstat
10:20:51 2 where name in (‘redo buffer allocation retries‘,‘redo entries‘);

NAME VALUE
---------------------------------------------------------------- ----------
redo entries 1125660
redo buffer allocation retries 9

Elapsed: 00:00:00.01
10:20:56 SYS@ test1>create table t1 as select * from scott.t1;

Table created.

Elapsed: 00:00:02.48
10:43:04 SYS@ test1>select a.name,b.value from v$statname a,v$mystat b
10:43:18 2 where a.STATISTIC#=b.STATISTIC# and a.name like ‘%redo size%‘;

NAME VALUE
---------------------------------------------------------------- ----------
redo size 13852204

Elapsed: 00:00:00.02
10:43:29 SYS@ test1>create index t1_id_index on t1(id);

Index created.

Elapsed: 00:00:17.44
10:44:08 SYS@ test1>select a.name,b.value from v$statname a,v$mystat b
10:44:20 2 where a.STATISTIC#=b.STATISTIC# and a.name like ‘%redo size%‘;

NAME VALUE
---------------------------------------------------------------- ----------
redo size 34085060

Elapsed: 00:00:00.01
10:44:24 SYS@ test1>

15:09:25 SYS@ test1 > select sid,event,seconds_in_wait,state
15:09:58 2 from v$session_wait
15:09:58 3 where event like ‘log%‘;

no rows selected

Elapsed: 00:00:00.00
15:09:59 SYS@ test1 >/

SID EVENT SECONDS_IN_WAIT STATE
---------- ---------------------------------------------------------------- --------------- -------------------
11 log file parallel write 0 WAITING

Elapsed: 00:00:00.00
15:10:02 SYS@ test1 >/

no rows selected

Elapsed: 00:00:00.00
15:10:03 SYS@ test1 >/

SID EVENT SECONDS_IN_WAIT STATE
---------- ---------------------------------------------------------------- --------------- -------------------
11 log file parallel write 0 WAITING
36 log buffer space 1 WAITING
37 log buffer space 1 WAITING

Elapsed: 00:00:00.00
15:10:04 SYS@ test1 >

解决方法：

可以调整log buffer 空间大小（不推荐），主要是调整lgwr的写出的速度

1、对于大的事务量，尽量采用分批提交
2、将redo 日志存放到磁盘I/O 最快的磁盘上（比如 采用RAID10方式）
3、增加日志组，增加日志组的大小
4、如果是归档模式，可再启用多个归档进程

二、java pool

在oracle 10g 采用内存自动管理，可动态分配java pool的大小

三、启动异步读写

异步： 当dbwr 写时，后台进程可以继续运行
同步： 当dbwr 写时，后台进程必须处于等待状态（串行方式）
10:44:24 SYS@ test1>show parameter syn

NAME TYPE VALUE
------------------------------------ ----------- ------------------------------
disk_asynch_io boolean TRUE
tape_asynch_io boolean TRUE
10:56:49 SYS@ test1>

------可以启用多个dbwr 进程（dbw0-dbw9）
10:57:37 SYS@ test1>show parameter writer

NAME TYPE VALUE
------------------------------------ ----------- ------------------------------
db_writer_processes integer 1
10:57:56 SYS@ test1>

----但和cpu个数有关，如果是多个cpu ，可以启动多个dbwr

db_writer_processes integer 1
10:57:56 SYS@ test1>show parameter cpu

NAME TYPE VALUE
------------------------------------ ----------- ------------------------------
cpu_count integer 1

调优5（SGA其他缓存区调整）,布布扣,bubuko.com

调优5（SGA其他缓存区调整）

标签：class   java   c   int   get   com   

原文地址：http://www.cnblogs.com/gispf/p/3725460.html