AIX

I have an unsupported legacy server:

# uname -a
AIX ibmms01 3 4 0054960A4C00

Following a period of poor performance, it was investigated using vmstat 10.
There upto 10 blocked kernel threads reported, and free memory was down to 123. The scan rate was high, 3000+ with lots of page in page out activity.

There was ample space on the file systems, the CPU was 60% idle.

The RAM was checked:

# lsps -a
Page Space Physical Volume Volume Group Size %Used Active Auto Type
hd6 hdisk0 rootvg 2272MB 50 yes yes lv
# lsdev -C | grep mem
mem0 Available 00-00 Memory
# lsattr -El mem0
size 2048 Total amount of physical memory in Mbytes False
goodsize 2048 Amount of usable physical memory in Mbytes False
#

This shows 2Gb RAM and 2 Gb swap space. Does the output of lsps -a show that only 50% of current available space is allocated for swap? If so how can the remaining 50% be allocated?

I've had a look in SMIT, but nothing seems obvious?

Thanks,

Jabberwocky

I don't use AIX and I do not understand the commands you used except for vmstat. But I can make a few general comments. It is not clear from your post what effect you would like to have on your system. Running out of swap would be a disaster. You would get messages like "can't fork", "out of memory", etc. Many programs do not handle errors like these very well. The box well might crash or lock up and a reboot would be needed to get things going again. Running out of swap means that you are out of virtual memory. Running out of physical memory is not good but perhaps not a disaster. If vmstat showed you a scan rate of 3000+, you clearly had been out of physical memory for some time. Generally, the box can run in this state, but with a performance impact. If you want to check your swap to ensure that were not close to a disasterous out-of-swap condition, that is a great idea, and I hope some AIX expert comes along who knows those other commands you used. But if you want to address that performance problem, you will need more memory. You may be able to make a modest impact on the severity of the performance problem by distributing your swap across more disks. But adding more memory can make the performance problem go away.

Excuse my previous vagueness.
My point is the output of lsps -a gives:

# lsps -a
Page Space Physical Volume Volume Group Size %Used Active Auto Type
hd6 hdisk0 rootvg 2272MB 50 yes yes lv

The %Used field value is 50 - does that mean that it is possible to
free up another 2272Mb of swap space on this AIX box?
If so, how?
I can see the swap on the root volume group in smit, but it is not clear
how to allocate the rest (if any exists) of the physical space on the
device.

Thanks,

Jabberwocky

You are only using 50% of your allocated swap space. In this case about 1.1G of swap.

Your real memory and swap are seperate devices. You have 2G of real memory AND 2.2G of swap space.

I think that has cleared it up lsps -a gives real time % usage of swap space.
I was originally thinking that the % Used column gave that amount of device- based swap space that has been allocated as in Solaris Disk Suite or Veritas Volume manager. The remaining space could then be allocated for use as swap space.
I now see that you mean this % Use figure relates to the current usage of swap space.

Thanks for your help.

Jabberwocky

Perhaps you haven't set any kernel tuning parameters. Per default MINPERM/MAXPERM is 20/80, try 10/40 and monitor the effect.

Also the MINFREE/MAXFREE parameters may have to be adjusted, but without more information i can't give you any detailed advice.

Use "/usr/samples/kernel/vmtune" (better consult the manpage before, this can send your machine to electronic nirwana) to set kernel tuning parameters.

You can use the vmtune utility to change parameters only until the next reboot. If you want your parameters to survive the reboot put them into an rc-file.

Btw. it is expected behavior that your machine doesn't have more meory free - AIX doesn't want memory to go wasted and will it assign to *any* task before leaving it unused.

bakunin

Thanks for your info. I have used lsattr -E -l sys0 command to look at the tunable kernel parameters. I cannot find any mention of minperm/maxperm nor minfree/maxfree in the output. Can these parameters only be accessed via kmtune? The box has live applications on it, so if invoking kmtune will affect the server in anyway, it won't be used.

The full output of lsattr -E -l sys0 gives:

# lsattr -E -l sys0
keylock normal State of system keylock at boot time False
maxbuf 20 Maximum number of pages in block I/O BUFFER CACHE True
maxmbuf 0 Maximum Kbytes of real memory allowed for MBUFS True
maxuproc 256 Maximum number of PROCESSES allowed per user True
autorestart false Automatically REBOOT system after a crash True
iostat true Continuously maintain DISK I/O history True
realmem 2097152 Amount of usable physical memory in Kbytes False
conslogin enable System Console Login False
fwversion IBM,M2P00279 Firmware version and revision levels False
maxpout 0 HIGH water mark for pending write I/Os per file True
minpout 0 LOW water mark for pending write I/Os per file True
fullcore false Enable full CORE dump True
pre430core false Use pre-430 style CORE dump True
rtasversion 1 Open Firmware RTAS version False
modelname IBM,7026-H80 Machine name False
systemid IBM,01654960A Hardware system identifier False
boottype disk N/A False
SW_dist_intr false Enable SW distribution of interrupts True
cpuguard disable CPU Guard True
#

Are there any parameters here that need their values changing?

Thanks,

Jabberwocky