Swap with 16 Gb of RAM

Dirk Wetter dirkw at rentec.com
Tue Aug 21 10:38:02 CDT 2001

Hi Vsevolod,

you should definitely go that far ;-)

I don't know what kernel are you running, but for your purposes
I wouldn't stick to the ones you're getting from RedHat, Debian, SuSE,

I am running a constantly growing cluster of 4GB SMP machines, and
dealing with VM issues under 2.4.x was a painful experience so far.
I am running 2.4.8pre4, newer kernels haven't been tested
yet in our enviroment, to 2.4.8 some things changed again in the VM,
2.4.9 is not a big step concerning the memory architecture.

The issue which you have sounds familiar to me: I had kswapd eating
all my CPU time, and it looked like the machine is heavily swapping, that was
as of 2.4.6 or .5. However, the machine didn't swap, it only looked like
it was. kswapd was trying to free pages, and it did it in every
memory zone. finally i got in touch with one of the VM guys, and
he came up with a patch, which solved our problem and was
later integrated into the stock kernel (2.4.7 i think).
So if you have 16GB of RAM and the early ineffective page scanner tries
to free pages (also in wrong zones, where there either is no shortage,
or it might not even help you), it takes a lot of time, 16GB by 4kB
are 4mio pages which have to be scanned.
 Also, 2.4.x kernels mark pages still as swapped, also if they are
in memory again (cached swapping). has the advantage, that if there's
a memory shortage, the page can be faster swapped out. So the numbers
for swap which you see might not be the right ones.  Recent kernels
however have in /proc/meminfo an entry "SwapCached" which gives you
more insight.

hope that helped a little bit,


 Dirk Wetter @ Renaissance Techn.
 mailto:<dirkw at rentec dot com>

 Tue, 21 Aug 2001, Vsevolod Ilyushchenko wrote:

> Hi,
> I have a Dell 8450 with 16 Gb of RAM used for running biological
> computations. Currently I have 3.5 Gb of swap, and sometimes it is almost
> completely used up even when only 4 Gb of RAM is used. Worse of all,
> sometimes kswapd takes up a lot of CPU time. Here is a recent free output:
>              total       used       free     shared    buffers     cached
> Mem:      16490952   16486332       4620          0        420   12197416
> -/+ buffers/cache:    4288496   12202456
> Swap:      3578664    2915448     663216
> I am trying to understand what is going on, but I do not know much about
> swap usage or how to debug it. I do not even know about the proper
> newsgroup/mailing list to ask this question, except the linux-kernel list.
> (I am not sure whether I should go that far. :)
> Thanks in advance for any ideas,
> Simon
> --
>  _________
> |       x |   Simon (Vsevolod ILyushchenko)   simonf at cshl.org
> | y = e   |
> |_________|   http://www.simonf.com          simonf at simonf.com
> 		Disclaimer: This is not me.
> 	This is just my mailer talking to your mailer...
> _______________________________________________
> Linux-PowerEdge mailing list
> Linux-PowerEdge at dell.com
> http://lists.us.dell.com/mailman/listinfo/linux-poweredge

More information about the Linux-PowerEdge mailing list