Closing.
]]>There's a wonderful article here http://www.peerwisdom.org/2013/04/03/la … rformance/
that explains what was happening: the VM was happily sending large packets to the host which subsequently was busy dealing with all the fragmenting and resending business, clogging one CPU core to the max and grinding the whole system to a screeching halt.
I just googled "LulzSec" and suddenly this kswapd problem appeared. Hope it's a coincidence :)
I still had 20 GiB free memory (no swap), but when I did as they recommend here
echo 3 > /proc/sys/vm/drop_caches
the problem disappeared. So I think it actually does something.
maybe you can try XFS not ext4 ?
I wouldn't recommend XFS unless you have a RAID array or an SSD. My experience with XFS was very much frustrating.
I have one 15K SAS direct attach drive, and it has a bad block now (and I can't re-allocate it using scu.d), and I've found out that XFS does not really support bad blocks. I can't fix the issue without re-booting! (that's how you solved problems back in the days of Windows 98).
Second, it's performance on such a setup is just horrible compared to ext3.
Moreover, my kernel version is plagued by an XFS bug when everything hangs suddenly. They call it something like a spontaneous bug that happens on "innocuous desktop workloads".
On the positive side, it does allow filesystem checks and de-fragmentations without re-booting, which ext4 doesn't.
]]>Hope it's not too late. And someone will be able to test my suggestion.
just noticed the same on RHEL6.3 and just wanted to say, maybe you can try XFS not ext4 ?
As I understand it's a booody ext with it's new features on ext4...
----------------- edit 1 ------------------
P.S. changing priority to lower it's not a solving issue, just fixing it for today...
----------------- edit 2 ------------------
also noticed on my RHEL box, that I also have big disk IO for these kswap0 issues, also jbd2 process is striking for io... check it with iotop.
This bug is not resolved yet.
]]>I then noticed that I hadn't updated my fstab to point to the correct swap partition (so I had no swap). I did that, then used swapon -a to turn it on, while running top in another terminal.
125MB was immediately written to the swap partition, and kswapd0's CPU usage stopped.
]]>Does anyone have any explanation as to what kswapd is doing while there is no swap configured? I have been able to 'fix' this by running echo 1 > /proc/sys/vm/drop_caches
]]>Did anyone find a fix?
]]>