2.4.20 & 2.4.22 paging out when it shouldn't
From: Brett (brettspamacct_at_fastclick.com)
Date: 09/30/03
- Previous message: Sergey S. Kostyliov: "Re: Call traces due to lost IRQ"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Date: Tue, 30 Sep 2003 10:34:18 -0700 To: linux-kernel@vger.kernel.org
We noticed that with linux kernel 2.4.20 and probably previous versions,
machines would at a certain time consistently go into paging overload
when we would coincidentially kill a few processes and start new ones.
The increased paging consistently coincides with the killing and
starting of processes at the same time every hour.
I am running sar, iostat and ps during the time that this happens and
what I see is sar showing pgpgout/s jump to 1000 or more for 30 seconds
with a corresponding increase in disk writing activity(iostat's
blk_writtn/s goes from about 15 to 5000) meanwhile the server is bogged
down, connections to the server time out and all hell breaks loose.
Also I see swap increasing. So I can only assume it's paging to disk.
Problem is there's around 500 megs of cache per top/sar info, we
shouldn't have to page.
So I added 500 megs of memory to give it a grand total of 1.5 gigs.
Same problem except the cache grew to 800 megs. So I did a swapoff -a.
Same problem except vmstat/sar show the swap is 0 yet sar reports high
pgpgout/s.
Next I upgraded to kernel version 2.4.22 and patched it with the latest
rmap(-rmap15k) patch, figuring this new VM would help. The cache became
a bit smaller. But it still paged out to disk.
I have gone over the linux-kernel mailing list archives and found others
who have run across a similar problem but there were no solid answers.
Someone recommended issuing this command as a workaround:
dd if=/dev/hda bs=8M count=$(awk '/MemTotal/ { printf "%d", $2/4096 }'
/proc/meminfo)
So I did that, kswapd took up 20-30% CPU, cache shot up. Then I killed
the process and the cache went down to 300 megs. So I figured I had
finally taken the disk cache down, freed up memory and it shouldn't
page. But it still paged.
Am I doing something wrong? It shouldn't page out to disk if I do
swapoff -a and have more than enough memory. Also, it should just kick
out the disk cache and use that for process pages instead of paging out
to disk, the disk cache isn't that valuable. It doesn't make sense so I
hope I'm doing something wrong. Any tips?
If anyone needs more information, please ask.
Thanks,
Brett
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
- Previous message: Sergey S. Kostyliov: "Re: Call traces due to lost IRQ"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Relevant Pages
|
|
