Found this discussion on a web search, so here is more confirmation that there's a problem in ps or the kernel:
# date; ps -O etime,start_time
Fri Jul 4 11:44:21 BST 2008
PID ELAPSED START S TTY TIME COMMAND
19589 02:39:16 09:05 S pts/1 00:00:00 bash
16022 49710-05:50:06 12:22 R pts/1 00:00:00 ps -O etime,start_time
# ls -ld /proc/19589
dr-xr-xr-x 3 root root 0 2008-07-04 08:26 /proc/19589
I did indeed log in and start the shell at 8:26, not 9:05.
That is, the ps results from the CPU reading is on this system (which has been up for 283 days) is about 38-39 minutes ahead of the system clock (
not the hwclock). It's not related to ntp which is running continuously, but rather the CPU drifting.
As you can see, the elapsed column is also unreliable (giving a wrapped result of many days, rather than the expected 0). If you want a reliable start time for the process, stat the /proc entry instead.
Debian 2.6.8-3-686-smp with procps/etch uptodate 1:3.2.7-3 It's not covered on the procps FAQ, but there are debian bug reports at
#408879 - /bin/ps: TIME drifted - Debian Bug report logs.