Hi, I have RHEL AS4 Update4 installed on VMESX3.02. I got the following error from ntpd daemon repeatedly.

Jan 1 22:01:12 RHEL4-001 ntpd[3066]: frequency error -504 PPM exceeds tolerance 500 PPM.
....

I got the error in the last two weeks and occured again today. Is it related VMware. Any idea about how to get the ntpd error disappeared, Please help.

I don't know why, but ntpd just doesn't work well under VMWare. I suspect the problem is that the guest machines don't have the privs to modify the clock in the hypervisor. The hypervisor is getting the interrupts from the clock, and ntpd modifies the kernel's estimate of the clock frequency to make it more accurate.

I think the solution is to get an NTP daemon that runs directly under the hypervisor. I have never heard of such a thing, but I don't see why it couldn't be written.


Jeff Silverman

There are a couple of options you can try. There are described in vmware and Red Hat's knowledge bases.

What I found to work almost fine is setting clock=pit, disabling ntp on the guest, and enabling time sync with the host (vmware tools).

Sorry for not providing you with links

To JeffSilverman
==>
I confirmed my vmware hypervisor also has NTP daemon enabled and ntpq -p
shows me that vmware hypervisor is synchronized with NTP server. It means
VM guest OS and hypervisor should has same clock.However, It still has ntpd frequency error.

To avalonit
==>
We tried to stop ntpd service on guest OS and sync time by using vmware-tool. But We found that OS systime will become inaccuracy.
I wonder what the meaning of clock=pit is?

How much inaccurate? Did the time go forward?

The kernel option tells kernel what timer source to use to timekeeping.


http://communities.vmware.com/thread/61253

You need both measures. clock=pit and disable ntp on guest + enable time sync through vmware-tools

Hi, avalonit

Thanks for giving the vmware KB link. I want to try to add the following boot options for preventing clock running slowly and quickly. I will update the result after OS next time reboot.

clock=pit nosmp noapic nolapic

And Yesterday I also got the same ntpd frequency error from one physical server running oracle service.

Jan 5 17:28:22 DB001 ntpd[13228]: frequency error -504 PPM exceeds tolerance 500 PPM
Jan 5 17:29:27 DB001 ntpd[13228]: frequency error -504 PPM exceeds tolerance 500 PPM
Jan 5 17:31:35 DB001 ntpd[13228]: frequency error -504 PPM exceeds tolerance 500 PPM

==> I found the difference between OS time and hardware clock always be 5s.
# date;clock
Mon Jan 5 17:44:01 JST 2009
Mon Jan 5 17:43:56 2009 -0.644794 seconds

I wonder whether [hwclock --systohc] can disappear the error as above.
If not, what should I do For example, restart ntpd service after [OS=>HW_Clock] sync.

I also worry that whether hw clock modification will cause oracle service down if I make
hw clock 5 second quickly.

If you have any idea, pls help me.

Thank you.

I see some pointers http://www.ece.udel.edu/~mills/ntp/html/debug.html

I don't *think* you should worry about that. But you should check if the hw clock is ever synced. On my laptop it was not and I found that by looking at the info displayed during shutdown. So I ended up setting the hwclock option --directisa. You could check if hwclock works without a reboot using your command from above.

As well you could check which timer source is your kernel using and possibly force another one.

I *think* changing *hw clock* should not cause any harm to oracle. But I can expect anything from it .
btw changing *system* time forward is much safer than backwards.

UPDATE:

http://kb.vmware.com/selfservice/mic...rnalId=1006427