[gentoo-user] ntp large drift (frequency error, frequent time resets)

[gentoo-user] ntp large drift (frequency error, frequent time resets)

[Alexander Puchmayr]

Hi there!

I think I have a problem with the system time, which is considerable too slow. 
It looses about 3 seconds every 20 minutes (i.e. ~10 secs/hour or 4 minutes 
per day). This seems to be too much for ntp to compensate.

When I start ntp, I get frequent "time reset" messages in the log:

2009-10-25T03:38:35.738758+01:00 poseidon ntpd[5227]: synchronized to 
131.130.251.107, stratum 1
2009-10-25T03:47:11.691165+01:00 poseidon ntpd[5227]: time reset +2.952038 s
2009-10-25T03:51:15.669040+01:00 poseidon ntpd[5227]: synchronized to 
LOCAL(0), stratum 10
2009-10-25T03:51:41.690598+01:00 poseidon ntpd[5227]: synchronized to 
131.130.251.107, stratum 1
2009-10-25T04:06:49.633882+01:00 poseidon ntpd[5227]: time reset +2.943313 s
2009-10-25T04:10:14.612358+01:00 poseidon ntpd[5227]: synchronized to 
LOCAL(0), stratum 10
2009-10-25T04:11:20.612348+01:00 poseidon ntpd[5227]: synchronized to 
131.130.251.107, stratum 1

I've also tried to start ntpd with "--slew" option, which results in

2009-10-26T08:36:20.171483+01:00 poseidon ntpd[21666]: frequency error 733 PPM 
exceeds tolerance 500 PPM
2009-10-26T08:40:25.167171+01:00 poseidon ntpd[21666]: frequency error 737 PPM 
exceeds tolerance 500 PPM
2009-10-26T08:47:05.169388+01:00 poseidon ntpd[21666]: synchronized to 
78.41.115.216, stratum 2
2009-10-26T08:47:05.169492+01:00 poseidon ntpd[21666]: frequency error 740 PPM 
exceeds tolerance 500 PPM
2009-10-26T08:47:30.168263+01:00 poseidon ntpd[21666]: frequency error 740 PPM 
exceeds tolerance 500 PPM
2009-10-26T08:50:42.168643+01:00 poseidon ntpd[21666]: frequency error 742 PPM 
exceeds tolerance 500 PPM
2009-10-26T08:58:09.172397+01:00 poseidon ntpd[21666]: synchronized to 
80.64.132.152, stratum 2
...

The next thing I've tried is to change the internal clocksource. The original 
setting was

cat /sys/devices/system/clocksource/clocksource0/current_clocksource
hpet

I've changed it to "acpi_pm" using 

echo "acpi_pm" 
>/sys/devices/system/clocksource/clocksource0/current_clocksource

but this doesn't seem to have any effect on the drift.

I'd like to use this machine as master time server for my local network, but 
this is useless with such problems.

Thanks in advance for any suggestions
	Alex

ps: ntp.conf contains (comments stripped)
server pool.ntp.org
server 0.gentoo.pool.ntp.org
server 1.gentoo.pool.ntp.org
server 2.gentoo.pool.ntp.org
server 3.gentoo.pool.ntp.org
driftfile       /var/lib/ntp/ntp.drift
restrict default nomodify nopeer
restrict 127.0.0.1
restrict 192.168.0.0 mask 255.255.0.0 nomodify nopeer notrap

Re: [gentoo-user] ntp large drift (frequency error, frequent time resets)

[Jil Larner]

Hi,

Alexander Puchmayr a écrit :
> I think I have a problem with the system time, which is considerable too slow. 
> It looses about 3 seconds every 20 minutes (i.e. ~10 secs/hour or 4 minutes 
> per day). This seems to be too much for ntp to compensate.

Is it physical or virtual environment? If virtual, clock issues are
common and, in case of VMWare, their tools handle them because ntp
cannot. Still, never use a virtual machine as a time reference.

If it's physical, my two cents are clueless. Maybe you wish to reduce
the frequency of the clock; I know it's a kernel parameter, but don't
remember which one.

Best regards,
Jil

Re: [gentoo-user] ntp large drift (frequency error, frequent time resets)

[Alexander Puchmayr]

Am Montag 26 Oktober 2009 16:23:11 schrieb Jil Larner:
> Hi,
> 
> Alexander Puchmayr a écrit :
> > I think I have a problem with the system time, which is considerable too
> > slow. It looses about 3 seconds every 20 minutes (i.e. ~10 secs/hour or 4
> > minutes per day). This seems to be too much for ntp to compensate.
> 
> Is it physical or virtual environment? If virtual, clock issues are
> common and, in case of VMWare, their tools handle them because ntp
> cannot. Still, never use a virtual machine as a time reference.
> 
> If it's physical, my two cents are clueless. Maybe you wish to reduce
> the frequency of the clock; I know it's a kernel parameter, but don't
> remember which one.
> 

Its a physical environment, although the server has some KVMs running. 

Greetings,
	Alex

Re: [gentoo-user] ntp large drift (frequency error, frequent time resets) [SOLVED]

[Alexander Puchmayr]

Hi folks

The problem is solved now. Actually, a BIOS upgrade solved the problem. 

Details:
The board is a Asus A3M78-Pro with a AMD 5050e Athlon64 Dual core CPU. Bios 
version was V0902 with with it was originally shipped.
Asus bios notes for that board say that with Version 11xx a bug was fixed 
which causes Vista's internal system time go too slow. Bingo. Linux seemed to 
be affected as well, so after upgrading to latest 13xx version, ntp behaves as 
it should :-)

Greetings
    Alex