After seeing some weirdness with a process and having to reset my time every couple months, I started running a cronjob, which tells me every day how much my clock has skewed:

20120118.log:Jan 18 03:30:10 ntpd[8]: ntpd: time set -1.072508 s
20120119.log:Jan 19 03:30:10 ntpd[6]: ntpd: time set -1.081339 s
20120120.log:Jan 20 03:30:14 ntpd[5]: ntpd: time set -1.083743 s
20120121.log:Jan 21 03:30:11 ntpd[0]: ntpd: time set -1.079067 s
20120122.log:Jan 22 03:30:12 ntpd[3]: ntpd: time set -1.074889 s
20120123.log:Jan 23 03:30:14 ntpd[2]: ntpd: time set -1.081986 s
20120124.log:Jan 24 03:30:11 ntpd[0]: ntpd: time set -1.074891 s
20120125.log:Jan 25 03:30:14 ntpd[5]: ntpd: time set -1.071300 s
20120126.log:Jan 26 03:30:12 ntpd[6]: ntpd: time set -1.073442 s
20120127.log:Jan 27 03:30:13 ntpd[7]: ntpd: time set -1.076903 s
20120128.log:Jan 28 03:30:10 ntpd[0]: ntpd: time set -1.076255 s
20120129.log:Jan 29 03:30:08 ntpd[0]: ntpd: time set -1.086898 s
20120130.log:Jan 30 03:30:12 ntpd[8]: ntpd: time set -1.070327 s
20120131.log:Jan 31 03:30:12 ntpd[1]: ntpd: time set -1.080005 s
20120201.log:Feb  1 03:30:13 ntpd[1]: ntpd: time set -1.066941 s
20120202.log:Feb  2 03:30:14 ntpd[]: ntpd: time set -1.077482 s
20120203.log:Feb  3 03:30:14 ntpd[8]: ntpd: time set -1.071182 s
20120204.log:Feb  4 03:30:08 ntpd[6]: ntpd: time set -1.071026 s
20120205.log:Feb  5 03:30:11 ntpd[7]: ntpd: time set -1.064174 s
20120206.log:Feb  6 03:30:09 ntpd[8]: ntpd: time set -1.062301 s
20120207.log:Feb  7 03:30:10 ntpd[8]: ntpd: time set -1.083911 s
20120208.log:Feb  8 03:30:13 ntpd[5]: ntpd: time set -1.069693 s
20120209.log:Feb  9 03:30:09 ntpd[]: ntpd: time set -1.072981 s
20120210.log:Feb 10 03:30:11 ntpd[]: ntpd: time set -1.080916 s
20120211.log:Feb 11 03:30:14 ntpd[4]: ntpd: time set -1.092480 s
20120212.log:Feb 12 03:30:10 ntpd[]: ntpd: time set -1.081575 s
20120213.log:Feb 13 03:30:11 ntpd[7]: ntpd: time set -1.082324 s
20120214.log:Feb 14 03:30:10 ntpd[]: ntpd: time set -1.083865 s
20120215.log:Feb 15 03:30:10 ntpd[]: ntpd: time set -1.085643 s
20120216.log:Feb 16 03:30:12 ntpd[]: ntpd: time set -1.070344 s
20120217.log:Feb 17 03:30:12 ntpd[3]: ntpd: time set -0.981060 s
20120218.log:Feb 18 03:30:12 ntpd[0]: ntpd: time set -1.188038 s
20120219.log:Feb 19 03:30:10 ntpd[4]: ntpd: time set -1.073639 s
20120220.log:Feb 20 03:30:11 ntpd[6]: ntpd: time set -1.096658 s
20120221.log:Feb 21 03:30:11 ntpd[5]: ntpd: time set -1.100242 s
20120222.log:Feb 22 03:30:11 ntpd[4]: ntpd: time set -1.090188 s
20120223.log:Feb 23 03:30:10 ntpd[1]: ntpd: time set -1.098994 s
20120224.log:Feb 24 03:30:09 ntpd[]: ntpd: time set -1.099090 s
20120225.log:Feb 25 03:30:13 ntpd[]: ntpd: time set -1.104736 s
20120226.log:Feb 26 03:30:11 ntpd[]: ntpd: time set -1.104673 s
20120227.log:Feb 27 03:30:12 ntpd[]: ntpd: time set -1.095368 s
20120228.log:Feb 28 03:30:10 ntpd[]: ntpd: time set -1.095536 s
20120229.log:Feb 29 03:30:11 ntpd[]: ntpd: time set -1.103140 s
20120301.log:Mar  1 03:30:15 ntpd[]: ntpd: time set -1.098549 s
20120302.log:Mar  2 03:30:13 ntpd[]: ntpd: time set -1.087177 s
20120303.log:Mar  3 03:30:14 ntpd[]: ntpd: time set -1.087147 s
20120304.log:Mar  4 03:30:13 ntpd[1]: ntpd: time set -1.086802 s
20120305.log:Mar  5 03:30:12 ntpd[]: ntpd: time set -1.077618 s
20120306.log:Mar  6 03:30:15 ntpd[1]: ntpd: time set -1.093846 s
20120307.log:Mar  7 03:30:15 ntpd[]: ntpd: time set -1.086906 s
20120308.log:Mar  8 03:30:12 ntpd[]: ntpd: time set -1.076139 s
20120309.log:Mar  9 03:30:13 ntpd[]: ntpd: time set -1.080558 s
20120310.log:Mar 10 03:30:07 ntpd[9]: ntpd: time set -1.088078 s
20120311.log:Mar 11 03:30:14 ntpd[8]: ntpd: time set -1.046163 s
20120312.log:Mar 12 03:30:12 ntpd[8]: ntpd: time set -1.083857 s
20120313.log:Mar 13 03:30:13 ntpd[0]: ntpd: time set -1.125479 s
20120314.log:Mar 14 03:30:14 ntpd[3]: ntpd: time set -1.044395 s
20120315.log:Mar 15 03:30:11 ntpd[3]: ntpd: time set -1.086920 s
20120316.log:Mar 16 03:30:13 ntpd[4]: ntpd: time set -1.079328 s
20120317.log:Mar 17 03:30:13 ntpd[3]: ntpd: time set -1.072813 s
20120318.log:Mar 18 03:30:12 ntpd[]: ntpd: time set -1.086966 s
20120319.log:Mar 19 03:30:11 ntpd[]: ntpd: time set -1.079843 s
20120320.log:Mar 20 03:30:12 ntpd[]: ntpd: time set -1.080058 s
20120321.log:Mar 21 03:30:11 ntpd[]: ntpd: time set -1.086402 s
20120322.log:Mar 22 03:30:12 ntpd[]: ntpd: time set -1.098703 s
20120323.log:Mar 23 03:30:15 ntpd[]: ntpd: time set -1.100180 s
20120324.log:Mar 24 03:30:52 ntpd[8]: ntpd: time set -1.064483 s
20120325.log:Mar 25 03:30:11 ntpd[0]: ntpd: time set -0.425544 s
20120326.log:Mar 26 03:30:16 ntpd[]: ntpd: time set -1.070984 s
20120327.log:Mar 27 03:30:08 ntpd[]: ntpd: time set -1.038029 s
20120328.log:Mar 28 03:30:15 ntpd[4]: ntpd: time set -1.078124 s

Does anybody else have this? What might cause it?

Do you by any chance have something scheduled to run at 03:30 that changes the time?

Considering that ntpd (which should be running continuously) always logs this at 03:30 I would almost think that you have something croned that runs eg ntpdate at that time, which sets a time that differs from what ntpd thinks is correct...?

Just guesswork of course...

Yea, I run /usr/sbin/ntpd -q. NTPd doesn't run continuously:

$ ps aux | grep ntp
root       496  0.0  0.1   6452   620 pts/2    S+   09:07   0:00 grep -i --exclude-dir=.svn ntp
$ 

The slew seems to be happening naturally, which I correct once a day.

Run an ntp daemon that will correct the clock in small increments constantly so you won't get the one second jump anymore.

_________________
Paid support
How to ask for help
1. Give details of your problem
2. Post any errors
3. Post relevant logs.
4. Don't hide details i.e. your domain, it just makes things harder
5. Be polite or you'll be eaten by a grue

Just like on a physical hardware, you need to run ntpd for it to keep good time.

-Chris

Without an atomic clock, or some way of synchronising with an external one (say via ntp), the most accurate we can make any electronic clock or oscillator will be to around +/- 15 seconds per month. Things which vibrate or oscillate are affected by things like pressure or temperature.

Losing an entire second in a day falls outside this range but not by much, so it's not grossly unreasonable. Running ntpd as a daemon will help you out.

Unless you get an OCXO, which controls temperature (and pressure?) and Wikipedia says should be accurate to a few dozen ms a month.

_________________
Matt Nordhoff (aka Peng on IRC)

That's interesting info, thanks.

After posting my previous post I realised I was replying to something that was weeks old, sorry about that.