Hi,

I reported a bug with 2.6.12.1-linode13 oom killer some time ago which I could solve by reverting to 2.6.10-linode12. I saw 2.6.12.3-linode14 today and gave it a try.

I had the same problem: the oom repeatedly killed some tasks although the system barely hit the swap. This time I found the culprit: I changed the default in /proc/sys/vm/swappiness to "1" (instead of 60). It seems there's a kernel bug triggered by this value. Setting swappiness to either 0 or 100 works ok (I can trigger the oom with 100% probabilty with one of my cgi scripts if set to "1" and couldn't trigger it with either 0 or 100).

Hope this can help someone hitting the same problem.