Hello,

If MAXPERM%=25 (i.e. 25% of Real Memory) on an AIX box - is it possible
to have currently in 'topas' "Noncomp 40%" (i.e. file chache consume
40% of RAM - 15 % above MAXPERM%) and increasing SWAP ussage at the
same time?

As I know with the current situation (MAXPERM%=25 and "Noncomp 40%")
AIX VMM will steal only file pages and as file pages are not swaped to
the paging space no SWAP space will be utilized - until "Noncomp"
memory fail bellow MAXPERM% (i.e. 25%).

Is it possible to have Paging (SWAP) space allocation while "Noncomp"
did not go bellow MAXPERM% at all - i.e. VMM swaps out program pages
when file pages in memory are much highier than MAXPERM%?


Thanks in advance,
George

maxperm% only comes into fruition once you hit minfree ... unless you
set strict_maxperm=1

HTH
Mark Taylor

Thanks, but strict_maxperm sets the hard limit which is not the target.
My question is why NUMPERM% is above MAXPERM% when there is no free
memory (so we have increasing SWAP usage). In this case as the server
is running out of free memory the VMM manager should steal only pages
from the file cache until it fall bellow MAXPERM% - which does not
happen.

George

Ah, that was relevant .. so you have already reached minperm ?

Can you post some vmstat 1 20 data from your system under load ... also
vmstat -v (if running 5.3) or vmo -a (5.2) or vmtune (5.1) ...

Make sure you have set minfreee high enough to satsfy kernel requests
for frames while the scanner catches up as if the freelist drops to
zero then you will still page out even if numperm is above maxperm.

Rgds
Mark Taylor

also, you should check you paging space allocation policy to see if
someone has changed it from "defered" to "late" or "early" ... check
out the docs

ref:
http://publib.boulder.ibm.com/infoce...d/memperf7.htm

Rgds
Mark Taylor

Hello,
The behaviour you describe is totally abnormal and seems to be related
with a problem I met a few months ago.
To be completely sure that it is the case, can you provide me the
following elements :
- AIX version with the maintenance level you are running.
- Is the box a Regatta LPAR ?
- How much memory is defined into that LPAR ?
- Check if the memory would be spreaded onto minimum 2 MCMs.
..
Thanks in advance.
..
Patrice.

Hello,
The behaviour you describe is totally abnormal and seems to be related
with a problem I met a few months ago.
To be completely sure that it is the case, can you provide me the
following elements :
- AIX version with the maintenance level you are running.
- Is the box a Regatta LPAR ?
- How much memory is defined into that LPAR ?
- Check if the memory would be spreaded onto minimum 2 MCMs.
..
Thanks in advance.
..
Patrice.

Hello,

This realy seems abnormal to me as well and this is why I posted this
message. This is an AIX 5200-06 LPAR with 16 GB of RAM (although I
don't know if it is Regata - I even don't know what Regata is in this
terms). Unfortunately I also don't know what means to spead memory on 2
MCM - I have good comprehension of AIX, but not on LPAR-s particularly.
On the box MINPERM%=10 , MAXPERM=15% and I see constant NUMPERM% ~ 30%
at the same time a pagging occurs (so code pages are sent to the SWAP)

Regards,
George

Can you post some data as requested above i.e.

vmo -a
vmstat 1 20
lslpp -l bos.perf.tune
lslpp -l bos.mp*
ls -l /unix
bootinfo -K

Rgds
Mark Taylor

Hello,

A regatta is a partitionnable server.
If your partition is on a Regatta, which is perhaps the case, there is
a new tunable parameter called memory_affinity which came with AIX 5.2
ML 4 (I'm not completely sure on the maintenance level).
IBM doesn't communicate a lot on this parameter but it seems that it is
not working so good.
You can see with vmo -a that this parameter is set by default to 1,
which means that the VMM tries to work better with memory located on
the same MCM (memory block). If it can't and if memory_affinity is set
to 1, VMM reacts as prefering working with paging space than with RAM
located on another MCM !!!
To disable this behaviour and continue to work as before, set
memory_affinity to 0 as follows :
vmo -r -o memory_affinity=0 and reboot the box. Check that the
parameter is correctly set after reboot.
You'll be probably able to see your box work as before.
Hope this helps.
Patrice