Re: oom with 2.6.11

[Christian Kujau <evil@g-house.de>]

hi again,

i had to wait for my pppoe session to be terminated by the remote peer
[1], and now it happened again with 2.6.11-rc5-bk2:

http://nerdbynature.de/bits/sheep/2.6.11/oom/oom_2.6.11-rc5-bk2_2.txt
http://nerdbynature.de/bits/sheep/2.6.11/oom/lsmod_2.6.11-rc5-bk2
http://nerdbynature.de/bits/sheep/2.6.11/oom/config-2.6.11-rc5-bk2.gz

i wanted to see the application chewing up the most RAM so i was running
"ps aux --sort=-vsz,-rss | head -n1" every 5sec and wrote it's output to a
file:

http://nerdbynature.de/bits/sheep/2.6.11/oom/daily_stats-2.6.11-rc5-bk2.log.gz

but even when almost all swap is used up, mysqld ist still no#1 with
131928 KB VSZ, as always. still suspecting something wrong with pppd, the
binary istself does *not* show up on the top-ten-memory-pigs.

real memory is almost always used up, but that's ok.
i stripped it down to show only the timestamp and the "Swap: .. used" numbers:

http://nerdbynature.de/bits/sheep/2.6.11/oom/daily_stats-2.6.11-rc5-bk2_stripped.log.gz

as you can see, until "Sat Mar 12 01:05:51" 122MB swap was used, from then
on swap-usage goes up until "Sat Mar 12 01:49:45", when all swap is in
use. the system is unable to keep up with writing memory usage every 5
seconds, the next entry is from 12 "01:56:24" - 6 minutes after the first
OOM message (i'm sure someone could feed this to gnuplot and make nice
bars out of it - i can't)

where else could i look to actually *see* where the memory goes to?

somehow lost in kernel versions,
Christian.

[1] "remote peer" aka "ISP". is there a way to "simulate" a "LCP
terminated by peer" locally? otherwise i really have to wait 24h to
trigger the bug.
-- 
BOFH excuse #373:

Suspicious pointer corrupted virtual machine