[lorenc.saraci]

I have a performance problem with one of my p570 servers (AIX5L 5.3). It shows the %busy=100 for hdisk0. I see that the used pagin space is not more than 15%. The nmon tool is showing a high value of Page Faults (around 35k).
How can solve this problem.

Many Thanks
Lorenc

[duke900ssd]

Which problem?
The high page faults?
The "performance problem"?
The 100% busy disk?
The paging not more than 15%

I guess you mean the 100% busy disk.
Try iostat and sar command's man pages for info on these commands.
If you have IBM cover raise a perfpmr.

[cdelgadop]

Hi

Could you please post the output of the following command:

vmo -a | grep -Ei "maxperm|minperm|maxclient"

Thanks

[duke900ssd]

Thrashing?

[lorenc.saraci]

Originally Posted by cdelgadop

Hi

Could you please post the output of the following command:

vmo -a | grep -Ei "maxperm|minperm|maxclient"

Thanks

Hi,

I had two problems the page faults was too high and the disk% busy was almost 100% all the time. I splited the paging logical volume in two hdd and now disk% busy is 20% but the page faults is still too hight.

The output for vmo -a | grep -Ei "maxperm|minperm|maxclient"
is as below

maclient%=80
minperm=384487
minperm%=20

Thanks
Lorenc

[cdelgadop]

Hi

Yeap. Thrashing

UNIX Memory Management
What is disk thrashing? - A Word Definition From the Webopedia Computer Dictionary

Lorenc. Depending on your AIX version i would reconfigured those values for maxclient/maxperm/minperm. One of the beauties of AIX is the VMM (Virtual Memory Manager), the SW layed in charge of managing all stuff related to paging space and real memory.

When having systems running programs that consume lots of real memory and do their own caching like RDMBS (Oracle, DB2, Sybase) it's a good practice to change the way VMM works no no double buffering (at DB level and OS level) is done.

I usually set those parameters with the following values:

maxclient%=08
maxperm%=08
minperm%=03

And it work for me. There're so many discussions about right values and latest Maintenance Levels (or Tech levels) recommends another numbers. If you're running SAP systems these above are the recommended values.

http://www.circle4.com/jaqui/eserver...hIOOandVMO.pdf

Hope this helps

[cdelgadop]

Hi

Sorry. about page faults too high. When CPU looks for a real memory page of data and that one is not there on real RAM then a page faults occurs. Then VMM will look for the data and ask disk subsystem to look for data, once disk susbsystem has found and retrieve data, VMM will place it in a free page on Real memory (it will then use the lrud program to download data from the least recently used pages of real memory to based paging space free page) and place data on RAM. As you can see if your system has lots of page faults data might not be get enough RAM memory to be placed on and disk access (that's by far more costly than RAM access) has to be done to look and retrieve data and to relocate frames or memory pages to paging space.

Your goal is to free as much real memory as possible, either by optimizing your VMM or by tuning your system to free as much RAM as possible. i always try to drop services not needed and tune java programs.

Hope this helps

[ross.mather]

Originally Posted by cdelgadop

I usually set those parameters with the following values:

maxclient%=08
maxperm%=08
minperm%=03

And it work for me. There're so many discussions about right values and latest Maintenance Levels (or Tech levels) recommends another numbers. If you're running SAP systems these above are the recommended values.

Hope this helps

Watch this advice please as the Memory Management recommendations have changed with AIX 5.3. SAP now recommends the minperm% at 20 and maxperm / maxclient at 80%. I know because I just did a big SAP rollout. There are significant differences in the memory management model used by 5.3 from earlier versions so old advice may no longer be relevant for your application.

What data is on the heavily loaded disk? That will help you understand which bit of your application is accesiing heavily. If this is a database its often due to a poorly written query - which loops through large tables and doesn't have the memory to cope with having that table in memory. Often indexing a couple of tables and optimising the query will solve the problem much btter than trying to tune the VMM.