How to reliably get timestamp at which the system booted

[I caught some misconceptions here which I think will this post clear off eventually]

There should not be any difference since both refer to /var/run/utmp file, which has its own format to store the records. If at all there is any difference, then your utmp file is busted. uptime shows the amount of time that has passed since the system has been booted or how long the system has been running. It does not tell you the system clock or system boot time . System boot time information is stored /var/run/wtmp file.

[centos@centos temp]$ date; uptime; who -b
Fri Dec  9 20:41:40 IST 2011
 20:41:40 up  1:32,  2 users,  load average: 0.50, 0.37, 0.29
         system boot  2011-12-09 19:11

uptime refers as well /proc/uptime, which essentially keeps the counters in kernel.

[centos@centos temp]$ sleep 1; cat /proc/uptime; uptime; sleep 5; cat /proc/uptime ; uptime
5914.79 5271.83
 20:47:39 up  1:38,  2 users,  load average: 0.29, 0.31, 0.27
5920.07 5276.80
 20:47:44 up  1:38,  2 users,  load average: 0.56, 0.36, 0.29

/var/run/wtmp is referred by last/lastb commands. who & w refers /var/run/utmp file. last reboot will show a log of all reboots since the log file was created.

Additionally, if you are having /proc filesystem, then tool such as procinfo can give you bootup time as well.

Example:

bash$ procinfo | grep Bootup
Bootup: Wed Mar 21 15:15:50 2001 Load average: 0.04 0.21 0.34 3/47 6829

On any POSIX-compliant system, you can use the etime column of ps.

LC_ALL=POSIX ps -o etime= -p $PID

The output is broken down into days, hours, minutes and seconds with the syntax [[dd-]hh:]mm:ss. You can work it back into a number of seconds with simple arithmetic:

t=$(LC_ALL=POSIX ps -o etime= -p $PID)
d=0 h=0
case $t in *-*) d=$((0 + ${t%%-*})); t=${t#*-};; esac
case $t in *:*:*) h=$((0 + ${t%%:*})); t=${t#*:};; esac
s=$((10#$d*86400 + 10#$h*3600 + 10#${t%%:*}*60 + 10#${t#*:}))