求教：AIX 6.1 11GR2 大量asynch descriptor resize等待！

hctech

求教：AIX 6.1 11GR2 大量asynch descriptor resize等待！

对aix的aio不太熟悉，不知需要提供哪些信息。

附上AWR报告

hctech

查询了一下系统参数：
aio_maxservers=30
aio_minservers=3

现在修改为
aio_maxservers=700
aio_minservers=700

情况有所缓解，在awr中的top5已看不到这个事件，虽然还是有wait~~

Maclean Liu(刘相兵

Patch 12412983: PROCESS HUNG ON ASYNCH DESCRIPTOR RESIZE AFTER UPGRADE TO 11.2.0.2
包括覆盖 skgfr.o 和 ksfd.o

就MOS上的反应来看 Patch 12412983对 该 问题的fix力度更大一些，可能涉及一些底层行为的改变。且这个one -off patch有11.2.0.3的对应版本。




Patch 9829397: ASYNC DESCRIPTOR RESIZE 仅有base release 11.2.0.2的版本 。


建议可能的话 APPLY Patch 12412983

hctech

刘总帮提供一下Patch 9829397  吧~~我没有账号~~~感激不尽啊~！！

hctech

Patch 9829397  
Patch:12412983

到底应该是哪个patch呀？

hctech

如果不升级或打PATCH，

按照 BUG 9829397 里描述解决方式：set DISK_ASYNCH_IO = false 这样处理的话会对系统有何影响？是否会导致IO性能下降，谢谢各位大大！

mfkqwyc86

Things to Consider Before Upgrading to 11.2.0.3 to Avoid Poor Performance or Wrong Results [ID 1392633.1]



Patch:12412983 AIX: "asynch descriptor resize" wait not necessary on AIX

Maclean Liu(刘相兵

ODM DATA:

11.2.0.2 on AIX-Based Systems (64-bit)  RAC=NO

Elapsed:                 60.03 (mins)                  
DB Time:                 55.20 (mins)         


Top 5 Timed Foreground Events

Event        Waits        Time(s)        Avg wait (ms)        % DB time        Wait Class
DB CPU                 2,725                 82.27         
log file sync        9,026        13        1        0.40        Commit
SQL*Net message to client        6,352,900        6        0        0.19        Network
asynch descriptor resize        2,526,757        3        0        0.08        Other


asynch descriptor resize  ==> other class   60分钟内 总等待时间 3s 不是性能的瓶颈


ODM FINDING:


EVENT: 'asynch descriptor resize'

Purpose
To understand this undocumented wait event 'asynch descriptor resize'
EVENT: 'asynch descriptor resize'
Wait event 'asynch descriptor resize' is set when the number of asynchronous descriptors reserved inside the OS kernel has to be readjusted.

It is signaled when the number of asynchronous I/O's submitted by a process has to be increased. Many Unix Kernels (for example: Linux kernel) do not allow the limit to be increased when there are outstanding I/O's; all outstanding I/O's must be resolved before the limit is increased. This event is shown when the kernel wants to increase the limit and is waiting for all the outstanding I/O's to be resolved so that the increase can be implemented.

If you see this wait event often, it might be a good idea to install the fix for Bug: 9829397 ASYNC DESCRIPTOR RESIZE.



Patch 9829397: ASYNC DESCRIPTOR RESIZE FOR Oracle 11.2.0.2.0  更新了 ksfd.o 文件


建议你 安装 Patch 9829397  或者升级到 11.2.0.3 + latest PSU