2.6.14 kswapd eating too much CPU

From: Jan Kasprzak
Date: Tue Nov 22 2005 - 07:59:46 EST


Hello,

I have noticed that on my system kswapd eats too much CPU time every two
hours or so. This started when I upgraded this server to 2.6.14.2 (was 2.6.13.2
before), and added another 4 GB of memory (to the total of 8GB). The server
runs mainly FTP load (proftpd w/ sendfile()), with some other apps (Apache,
Qmail, etc). It is dual Opteron (Tyan S2882 board) with 8 ATA disks
on 3ware 7506 controller with SW raid volumes (RAID-1 and RAID-5),
with XFS and ext3 filesystems.

Here is the top(1) output when the kswapd problem occurs:

top - 13:45:34 up 8 days, 15:49, 2 users, load average: 3.37, 3.49, 3.44
Tasks: 325 total, 4 running, 321 sleeping, 0 stopped, 0 zombie
Cpu(s): 0.5% us, 75.6% sy, 0.8% ni, 20.3% id, 1.3% wa, 0.2% hi, 1.3% si
Mem: 8174528k total, 8151076k used, 23452k free, 49960k buffers
Swap: 14651256k total, 640k used, 14650616k free, 7562572k cached

PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
16 root 15 0 0 0 0 R 96.4 0.0 1542:27 kswapd0
15 root 15 0 0 0 0 D 52.0 0.0 833:48.37 kswapd1
1056 rsyncd 35 19 11196 2608 928 R 3.0 0.0 0:42.29 rsync
2082 named 18 0 72816 17m 1888 S 0.7 0.2 47:47.55 named
18484 ftp 20 5 10028 1556 788 S 0.7 0.0 0:00.02 proftpd
26 root 10 -5 0 0 0 S 0.3 0.0 156:47.10 md5_raid5

It can be seen that one kswapd uses almost 100% of CPU,and the other one
about 50%. The MRTG graph of CPU usage can be seen at

http://www.linux.cz/stats/mrtg-rrd/cpu.html

(orange is the system+irq+softirq time, and red is non-idle (i.e.
user+sys+nice+irq+softirq, without idle and without iowait). The upgrade
of the kernel and memory has been done on Saturday, November 12th.

The peaks of the CPU system time happen every two horus or so. The peak
from ~5am is wider, because at that time various cron daily jobs are run
(updatedb, and so on).

Any clues on what is going on here? Thanks,

-Yenya

--
| Jan "Yenya" Kasprzak <kas at {fi.muni.cz - work | yenya.net - private}> |
| GPG: ID 1024/D3498839 Fingerprint 0D99A7FB206605D7 8B35FCDE05B18A5E |
| http://www.fi.muni.cz/~kas/ Journal: http://www.fi.muni.cz/~kas/blog/ |
> Specs are a basis for _talking_about_ things. But they are _not_ a basis <
> for implementing software. --Linus Torvalds <
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/