Just add the -p option to the command. So command
# vmo -p -o minfree=480 -o maxfree=512 -o minperm%=10 -o maxperm%=30
will both set the vmm options and make them permanent at the same time.
However, chances are that the old values from the old AIX and the old haredware are inapropriate for your new system. Because:
minfree/maxfree changed from a per system setting to a per cpu setting and probably is way too low anyway.
maxperm% must go with maxclient%
lru_file repage is available with current 5.3 TL and depending on your application your minperm%/maxperm% values might be not optimal
maxpageahead may be wrong if you changed from jfs to jfs2. In this case j2_maxPageReadAhead=32 is what you want.
to name just four things that come to mind without digging too much into the details.
Just have queary ? How can we know what maxperm/minperm or lru_file setting should be there for optimal performance. Is there any tool which can help to check it.
Having maxperm% and maxclient% around 30 is an old way of tuning. As shockneck said, lru_file_repage is being used since some time now and while using this, the maxperm% and maxclient% are usually kept high like 80 or 90 percent.
For an Oracle server you should have something like this:
Code:
vmo -p -o lru_file_repage=0
So it will only throw away cached files to get free pages, as long as it doesn't get below minperm when it will start to steal everything to get free memory. For that cause you might use these settings as well as a start:
Code:
maxperm%=80
maxclient%=80
minperm%=5
If you had paging, you can easy see it with "vmstat 1" for example (pi/po column). It takes effect immediately, but you might have to wait some time until the system calmed down and has cleaned up the memory.
Also you should have legacy async i/o (AIO) activated and monitor how many servers are being used with nmon maybe (not the number how many have been started) and how many async i/o requests appear (iostat -A).
You got to monitor your system if the changes bring a positive effect.
Also you might have a look with vmstat -v if any fsbufs, psbufs etc. are being blocked and increase the values which are effective after a remount of the FSes.
Thanks for the reply zaxxon and shockneck.. I appreciate it.
I have set the parameters like this as per your recommendations:
root@ebtstdb on / # vmo -p -o minfree=480 -o maxfree=512 -o minperm%=5 -o maxperm%=80 -o maxclient%=80
Setting maxfree to 512 in nextboot file
Setting minfree to 480 in nextboot file
Setting minperm% to 5 in nextboot file
Setting maxperm% to 80 in nextboot file
Setting maxclient% to 80 in nextboot file
Setting maxfree to 512
Setting minfree to 480
Setting minperm% to 5
Setting maxperm% to 80
Setting maxclient% to 80
I have not set the j2_maxPageReadAhead=32 or vmo -p -o lru_file_repage=0 yet..
Before i do that, could you analyze the output of vmstat and (iostat -A).
[...]
I have not set the j2_maxPageReadAhead=32 or vmo -p -o lru_file_repage=0 yet..
Before i do that, could you analyze the output of vmstat and (iostat -A).
[...]
Sorry, no. Changing the minperm/maxperm settings the way you did does REQUIRE to change the lru_file_repage algorithm also to really make sense. Set the lru_poll_intervall to 10 if it is at a lower value.
Then put some load on your system by running your application, watch memory usage and come back.
Last edited by shockneck; 03-03-2009 at 11:24 AM..
Good evening
I nned your help pls, In an unix server i want to connect to a remote oracle databse server by sqlplus.
I tried to find out the user/passwd and service name by env variable and all Ive got is this:
ORACLE_SID_REPCOL=SCL_REPCOL
ORACLE_SID=xmeta
ORACLE_SID_TOL=SCL_PROTOLCOL... (2 Replies)
Guy's
I'm looking for vmo parameters on AIX 6.1 to be for data base and application servers to enhance the performance
Is there any one has tested good vmo tuning and and felt a good performance on his server
Pls advice and provide.. :rolleyes: (1 Reply)
