【12c新特性】多LGWR进程SCALABLE LGWR “_use_single_log

SCALABLE LGWR是12cR1中引入的一个令人激动的特性，这是由于在OLTP环境中LGWR写日志往往成为系统的主要性能瓶颈，如果LGWR进程能像DBWR(DBW0~DBWn)那样多进程写出redo到LOGFILE那么就可能大幅释放OLTP的并发能力，增长Transcation系统的单位时间事务处理能力。

在12cR1 中真正用SCALABLE LGWR实现了这个目的，也可以俗称为多LGWR进程。

         select * from opt_12cR1 where name like '%log%'
_use_single_log_writer	ADAPTIVE	Use a single process for redo log writing
_max_outstanding_log_writes	2	Maximum number of outstanding redo log writes

SCALABLE LGWR主要受到隐藏参数_use_single_log_writer的控制，该参数默认值为ADAPTIVE 。

该参数主要有三个可选值 true， false， adaptive，默认值为ADAPTIVE。

对于ADAPTIVE 和False 如果CPU个数大于一个则会有多个lg0n进程
对于true 则不会生成多个lg0n进程，而如同12.1之前那样仅有单个LGWR

SQL> show parameter _use_single_log_writer

NAME                                 TYPE        VALUE
------------------------------------ ----------- ------------------------------
_use_single_log_writer               string      ADAPTIVE



[oracle@maclean1 ~]$ ps -ef|grep lg
grid      4344     1  0 08:07 ?        00:00:00 asm_lgwr_+ASM1
oracle   12628     1  0 08:48 ?        00:00:00 ora_lgwr_MAC_1
oracle   12636     1  0 08:48 ?        00:00:00 ora_lg00_MAC_1
oracle   12640     1  0 08:48 ?        00:00:00 ora_lg01_MAC_1
oracle   13206  7447  0 08:51 pts/2    00:00:00 grep lg

可以使用 10468 level 2 来trace adaptive scalable LGWR

[oracle@maclean1 trace]$ oerr ora 10468 
10468, 00000, "log writer debug module"
// *Document: NO
// *Cause:
// *Action: Set this event to the appropriate level for log writer debugging.
//

alter system set events '10468 trace name context  forever,level 2';

LGWR TRACE:
kcrfw_slave_adaptive_updatemode: time=426523948079110 scalable slave=1 arbiter=3 group0=10678 all=12733 delay=100 rw=98774 single=1973
48 scalable_nopipe=197548 scalable_pipe=108651 scalable=180439
kcrfw_slave_adaptive_savewritecounts: time=426523954275133 group0=10695 all=12752
kcrfw_slave_adaptive_savewritecounts: time=426523954662537 group0=10696 all=12753

CKPT TRACE:

*** 2014-12-07 10:52:21.528
kcrfw_slave_adaptive_saveredorate: time=426523941528521 curr=16649627696 prev=16635613056 rate=14014640 avg=14307212

*** 2014-12-07 10:52:24.553
kcrfw_slave_adaptive_saveredorate: time=426523944553556 curr=16664120996 prev=16649627696 rate=14493300 avg=14318490

实际测试可以发现仅在redo 生成率非常高的环境中SCALABLE LGWR 对于redo写出的吞吐量有所帮助，进而提高OLTP环境的TPS。

_use_single_log_writer = adaptive 2个LG slave进程：

Per Second	Per Transaction	Per Exec	Per Call
DB Time(s):	2.8	0.0	0.00	0.33
DB CPU(s):	2.6	0.0	0.00	0.31
Redo size (bytes):	8,180,730.6	545.6
Logical read (blocks):	46,382.1	3.1
Block changes:	60,219.5	4.0

Function Name	Reads: Data	Reqs per sec	Data per sec	Writes: Data	Reqs per sec	Data per sec	Waits: Count	Avg Tm(ms)
LGWR	1M	0.14	.004M	4.3G	29.80	16.16M	1785	79.10

_use_single_log_writer = true 使用single lgwr

Per Second	Per Transaction	Per Exec	Per Call
DB Time(s):	2.8	0.0	0.00	0.34
DB CPU(s):	2.6	0.0	0.00	0.32
Redo size (bytes):	8,155,843.5	545.0
Logical read (blocks):	46,550.1	3.1
Block changes:	60,036.7	4.0

Function Name	Reads: Data	Reqs per sec	Data per sec	Writes: Data	Reqs per sec	Data per sec	Waits: Count	Avg Tm(ms)
LGWR	1M	0.13	.003M	4.8G	25.49	16.141M	1611	95.97

【12c新特性】多LGWR进程SCALABLE LGWR “_use_single_log_writer”

Comment 取消回复