Please start using code tags for code, logs data etc. You got a PM with a guide how and when to use them.
The output of the svmon -p| awk ... line is missing which would be also interessting so we can see what processes have been forced to use slow Paging Space.
What type of application is running there? Some DB? If yes, what size of memory have you dedicated for it to use (like SGA size in Oracle for example)?
From the vmo output it seems your system had been tuned already the old way, which is still good.
Optional: The usual way these days is to set something like the following:
These settings will be made permanent so after a reboot, they are still there. You can do them online.
So if you wanna try them, I suggest a reboot after setting them though so your memory will be cleaned up and you can have a look if your box still tends to use Paging Space.
Seems you simply ignore my requests for code tags of any kind. If you continue, you will gather infraction points until your account is banned.
I assume virtual server means LPAR. There is no difference to set the parameters to a single running box.
I have checked the output you have given of vmo. I have written what needs to be changed, and you just post 2 of 4 I have posted. Those 2 allone without the other 2 make no sense.
I have written how to change them, you can copy&paste it.
90% is neither the default nor the maximum. You should check the man page for vmo to see which parameter is the current, default and so on.
Or you can use another switch like -L when calling vmo to have a detailed table.
A crude description of what will happen is:
If your percentage of RAM used as file cache is above 90%, it will only kick out these, to free up RAM for processes and their working data.
If your RAM is between the 5% and 90%, then it will usually just kick out cached files too, as long as they have not been repaged more often than computational pages, ie. the files have not been more used than some process/processing data.
If cached files fall below 5% then everything will be stolen to free up RAM to guarantee the minimum of a file cache to have the system not being slowed down by disk access for important files.
Read carefully again what I have posted and what you ask for. Everything you need is there. If you did not understand, you can ask, but please be as precise as possible and don't post half the commands that have been recommended.
I think it is time you understand some of the basics of AIX memory-housekeeping:
How swap/paging space and memory is related
Unix is a OS from times long ago and long ago memory was VERY precious. All Unix systems can use disk space to emulate memory (RAM). Every system has:
a "virtual memory" size, which is, roughly speaking, the installed physical RAM plus the size of all paging spaces.
a "real (physical) memory size", which is the size of installed RAM alone
a "swap size" (or "paging area" size), which is the size of all paging spaces summed up
Unix does not (really) differentiate between virtual and physical memory as far as applications are concerned. If an application asks for memory, it gets some, but the application has no control (in fact no knowledge) about where this memory comes from. It can come from the paging area or the real memory. (There are ways to circumvent this, but more later.)
The kernel itself manages where the memory is in fact taken from: first it gives out all the memory it has, because memory is very fast compared to disks. Only when real memory is exhausted it shovels the least-used memory pages to the swap ("swapping out") and makes room for more memory demands. At some point, even the least-used memory page is used again and then, the kernel looks for the least-used memory page(s) again, puts them in the swap, making room this way ("swapping out" again) and shovels the used pages back in ("swapping in"). The application will not notice all this, but the user will, because this swapping out/swapping in takes noticeable time. The more often it happens the slower the system will appear.
There is the command "vmstat" to see this happening. The columns "pi" (page in) and "po" (page out) are the number of memory pages swapped in and out in the sampling interval. Issue
to set the sampling interval to 1 second. (I don't know about you but my inner clock is counting in seconds, so this is a good starting point for me. Use longer intervals like 300 [=5 min] for long-term monitoring.) A "memory page" in AIX is usually 4k, so multiply with 4096 to get the sizes in bytes.
Do you need a bigger swap space?
Yes and no. No, because memory is already very cheap and you might want not to have to swap at all anyway. You might need to increase the RAM, use AME (a sort of RAM compression done transparently), use shared RAM (AMS) or whatever to make the paging space unnecessary. If you have a small unused paging space or a large unused paging space does not matter at all.
Yes, because paging space is your "plan B" if memory is exhausted. You have 32GB physical memory installed, so if you exhaust that at some time in the future you will do so probably by more than the mere 6GB you now have in contingency. Be aware, though, that this is pure conjecture. It might be that under very certain circumstances the 6GB might be optimal. Increasing the paging space adds nothing to the systems resources, it just pushes the point at which the system becomes absolutely unusable a bit farther back. To compare with a car: this is not a better working brake or a bigger motor, it is a bigger safety belt.
Notice the difference between swapping and swap usage
Notice that swap usage does not hurt, only the activity of swapping in or out memory pages does. What does that mean? Some process would be ready to run, but its memory pages are in the swap. In this case the process is stopped, its pages are swapped in by the kernel (and probably some other pages are swapped out instead) and only then it can start to run. In the meantime other processes could use the system (its processors, or whatever), but if all processes otherwise ready to run would be affected by swapping activity he machine effectively does nothing but wait. This is what makes it appear to be slow.
Now suppose there is a huge (in terms of memory) process, which only runs once a day: it might lurk in the swap, will be swapped in once a day then run for some time and be swapped out again. It would allocate a big amount of swap, but the actual swapping activity taking place would be minimal. This would not hurt at all in terms of perceptible performance.
With databases involved everything is different, of course
I said above that applications have no say where the memory they request comes from. This is in principle true, but there are exceptions to this rule. It is possible to explicitly ask for memory coming from RAM and should never be swapped out (so-called "pinned memory"), though this should not be done lightly (if everybody asks for physical memory it will be exhausted soon AND the swapping feature of Unix would effectively be disabled). Database software, though, often use memory to cache often-used contents of the database and it would make no sense whatsoever to have that swapped out. Therefore database software allocates "pinned memory" for some vital parts. In Oracle this is called "SGA" (System Global Area) and its size can be customized.
Talk with your DBA about how big it is and how big it needs to be (they always go for "as big as possible and bigger is better" ;-)). As Oracle does its own cache on dedicated Oracle servers you might want to cut back on so-called "file memory" (all the physical memory not immediately used is dedicated to file caching by the AIX kernel) and give the room made free this way to the database software.
I hope this helps.
bakunin
These 2 Users Gave Thanks to bakunin For This Post:
Location: on the road for work; home is private time
Posts: 456
Thanks Given: 10
Thanked 108 Times in 100 Posts
Zaxxon - is well ahead of me in terms of what needs to be said.
Before 1999 there were recommendations for having paging space at @X installed memory, and before 1996 (AIX v3) there were cases where 3X was recommended (i.e., when installed memory was <128 MByte - Mega, not Giga Byte).
Starting with AIX 4.3.3 - ML3 I believe, AIX introduced a new algorthim - differed paging space. From that time, the amount of paging space needed is drastically less.
On my main (personal) system, with 9G of memory I have 512M Byte of memory.
There will always be some paging space used - this occurs "during boot" as far as I can tell.
The difference between before AIX 4.3.3 ML3 and now is that paging space is only allocated when it is actually needed - NOW - while BEFORE, it was allocated (but not written!) when memory got malloc()ed (PSALLOC=early), or when the page was initially touched in physical memory (PSALLOC=late).
So, seeing paging space is 81% consumed - yes, increase it - make it a an additional space that is not automatically restarted. This is to prevent a "panic" like crash - avoid happenings like this whenever you can.
HOWEVER - the real problem is not your "lack of paging space" (although 9G (~= 25% of physical memory) does have a nice feel to it.) - your problem is that applications are needing it - as bakkin talks about.
You need to spend time figuring out WHY applications are needing the support of paging space. This is NOW - not 1988-1997 time frame when paging space was common - and actually needed because of the economics of purchasing memory.
In short, you need additional paging space because it will prevent a "panic crash" - but your problem is that your system needs paging space to prevent a crash. Find out why your system does not fit in 32G of memory. If there are valid reasons - increase the memory rather than the paging space.
My two cents - and hope it helps!
Michael
These 2 Users Gave Thanks to MichaelFelt For This Post:
Whilst perfoming some tests on lvm's I managed to crash our test box. No real problem as it is only used by our tech team.
however I would like to know why this was actually caused as the task being performed at the time was one which I though would not have any impact.
Using dd I was... (5 Replies)
Hi,
I have paging size 2048M showed from topas and 10240M showed from "lsps -a", can anyone tell what is the difference? and how to change the PAGING SIZE (showed in topas) to 8192M?
Can you please tell in detail step?
Thanks!
Victor
#topas
Topas Monitor for host: egsprc01dev ... (10 Replies)
Hello dear friends, We have VG filevg which consists of 2 PVs when I rechecked the VG there is no Pagingspace LV.. The VG is usually Highly loaded because much reads and writes.. Is this a must to create Paging space on the specified LV? (2 Replies)
Hello everyone
I have a doubt about how many paging space can have in the same disk.
lsps -a
Page Space Physical Volume Volume Group Size %Used Active Auto Type
paging00 hdisk0 rootvg 3072MB 1 yes yes lv
hd6 hdisk0 ... (4 Replies)
Hi
I have used gzip on AIX and the used paging space has jumped from 7% to 20%. The gzip process is finished since a long time. But the used paging space is still the same. How to release this space ? (1 Reply)
Hello everyone
I have 4g of paging space in my rootvg disk
I´m going to reduce them to 1gb in my rootvg disk and add 3gb of paging space on my san disk.
My rootvg disk is mirror.
My question is I can do this on line ?
and I can do with the mirror ? or I need to unmirror first my... (2 Replies)
sorry for this silly question, I am new to UNIX,
what is meant by paging space and what is its purpose?
what is also meant by hd6 paging space? (2 Replies)
Hello, we have a problem with lpar with AIX 5.3, the issue is that has high level paging space with: _sqlsrv2 and its incremented continously (1 Reply)
This is my first post, and I am new to the UNIX world. Hopefully this question won't be too lame.
I know that I can use topas to see the paging space used by some processes. I would like to script something that can add up the paging space used by process owned by or associated with an... (1 Reply)