Clock drift?

[guerby]

Hi,

Since my plug mount / by NFS on a server which has ntpd when I build software I get warning about files in the future, etc... I didn't install ntpd on the plug so I run ntpdate by hand from time to time.

However I noticed something funny:

gcc55 ~ # ntpdate pool.ntp.org
23 Apr 11:01:48 ntpdate[20022]: step time server 88.191.89.73 offset 28.554061 sec
gcc55 ~ # uptime
 11:01:54 up 14 days, 16:56,  2 users,  load average: 0.04, 0.32, 0.41

The drift is 2 seconds per day.

Boot, run ntpdate, wait a few days and do ntpdate again, could someone report too?

Thanks!

[rshitrit]

which kernel are u using?
do u run heavy load on the plug, or leave it "idle"?

[jmknapp]

I've been running ntpd and the drift file/var/lib/ntp/ntp.drift reports a drift of 27.4 ppm, which indeed is 2.37 sec/day.

[guerby]

kernel was compiled from Marvell git:

Linux version 2.6.29-rc7-00063-g9d1d4f9-dirty (root@gcc55) (gcc version 4.1.2 (Gentoo 4.1.2 p1.3)) #12 PREEMPT Tue Mar 24 19:01:56 CET 2009

Load is some idle (80% of the time) and otherwise heavy compilation. But change in load does not seem to affect the clock drift which is surprisingly constant.

[eranba]

The clock drift is a result of up to 30 ppm deviation(according to technical specification) of the main 25MHz system oscillator which derive the 200MHz core timer. 

[birdman]

If hardware clocks were accurate no-one would have written ntp.
30ppm isn't much.  Far better than my typing skills.