Identifying performance problems 3

[phorbiuz]

Hi

General performance questions if you don't mind regarding CPU and memory? There's a fair bit of info here to hopefully provide necessary detail. Anyway, AIX 5300-06-00-0000, running Oracle, with a little over 11Gb RAM and just shy of 9Gb paging, with 4x 750Mhz cpu's.

Svmon -G:

size inuse free pin virtual
memory 2883584 2882738 846 250487 3280826
pg space 2228224 916708

work pers clnt
pin 250487 0 0
in use 2751363 131375 0

Topas snapshot:

CPU Kernel 24.0
User 75.9
Wait 0.0
Idle 0.0

Network and Disk stats minimal

Memory Real,MB 11264
% Comp 96.5
% Noncomp 4.1
% Client 0.5

Paging Size,MB 8704
% Used 42.4
% Free 57.5

Paging Faults 5414
Steals 0
PgspIn 39
PgspOut 0
PageIn 498
PageOut 78
Sios 378

Now for the questions bit.....

To me, this system is obviously short on memory and CPU but I'm no performance expert. The svmon command says the system is using all available RAM plus 4Gb of paging, with topas showing 96.5% of that being %comp.

1st. Is 96.5% abnormally high or acceptable?
2nd. Paging used is in excess of 40%. Is it still considered too high once you get past about 20%?
3rd. Over 5400 paging faults seems excessive too, wouldn't you agree?

Anyone any comments?

Thanks

 

[Chapter11]

Looks like you're hurting on memory to me too.

What's your minperm% and maxperm%? Defaults are 20% and 80% with non-strict maxperm%. This means that AIX will be doing a lot of disk caching in RAM, and if you're intending to use that for Oracle, then you're double-dipping in memory (and therefore causing paging).

I prefer 5% and 15%, with a strict maxperm%. If you do that, step down the maxperm% 2-5% at a time; each time you step it down, CPUs will flood with "s" activity briefly. If you do it all at once, then you'll hammer the system with a flood of "s" activity for a very noticable amount of time.

 

[khalidaaa]

Have alook at this thread, there are alot of nice comments on tuning:

http://www.tek-tips.com/viewthread.cfm?qid=1302007&page=1

Regards,
Khalid

 

[phorbiuz]

Settings as requested:

vmo -a | egrep "maxperm|minperm|maxclient|lru"
lru_file_repage = 0
lru_poll_interval = 10
lrubucket = 131072
maxclient% = 90
maxperm = 2429682
maxperm% = 90
minperm = 80989
minperm% = 3
strict_maxclient = 1
strict_maxperm = 1

This system seems to be set at 3 and 90%, although I have no idea why and who set it.....

 

[bdw238]

Hello,

Just over 6 months ago we had a few pseries servers with Oracle 9 and 10 rebooting themself's and asked IBM( via Aix Software support) to looking into. They recommanded the following settings, which seemed to do the trick

lru_file_repage = 0
strict_maxperm = 0
strict_maxclient = 1
maxperm% = maxclient% = <high percentage>
minperm%=<low percentage>
lru_poll_interval=10



Looking at your last post, you have this settings applied, so virutal memory settings are not the issue. The orginal post you sent seemed be a bit General and focused on high memory usage been a problem, which can be a bit of red Herring ( distraction from an original objective)!

1.
What you need to do is use nmon, look at the VMtune graph on the memuse sheet. Look for:

a) Is the %comp Constantly greater then %maxperm?
b) is the %numperm constantly below/near to %minperm?

In answer is yes to both, then there is a memory issue, as virtual memory manager is forced to do page anthing at will. You then need to look at in order to fix the problem:

1. Look at CPU usage and Run queue. If they are high, they may need more cpu's as system is having keep a blocked process(s) in memory, which thus uses more memory!

Thus, the solution will be either

1. Improve the application on the box if possible
2. Buy more cpu and/or memory.


Brian

 

[costiles]

I use ps auxw|pg to look at the processes using the most cpu - and ps auxw|sort -k5rn|pg to find the processes using the most memory - then investigate those processes. note the flags do not have the (-) sign on the flags.