NTP (05)

NTP Still Wrong

Wednesday 23 April, 2008 - 23:36

Once more upon the breech! In Installing and Configuring NTP on VMware ESX Server, I found the following commands:

hwclock --systohc
watch "ntpq -p"

The hwclock --systohc is supposed to set the hardware clock based upon the system clock periodically.

The interesting command was ntpq -p. I could watch the offset increase up to 10 seconds within a few minutes.

I checked the vmware log for clues:

Msg_Hint: msg.hostWin32.cpuSpeedMismatch (sent)

VMware Workstation has measured your CPU speed to be 1881 MHz, but Windows reports that it is 2001 MHz. This may mean that your computer has a power-saving feature that varies the processor speed. As a result, the clock in your virtual machine may run too fast or too slow.

For a workaround, please refer to the VMware knowledge base article at:

http://www.vmware.com/info?id=97

I had noted this problem back in Basic Cluster but ignored it.

I created c:\Program Files\vmware\Workstation\config.ini with:

host.cpukHz = "2001000"
host.noTSC = "TRUE"
ptsc.noTSC = "TRUE"

I restarted the VMware Authorisation service and started the YAOCM team. When I ran ntpq -p after a few minutes, I got:

remote refid st t when poll reach delay offset jitter
central.yaocm.i LOCAL(0) 11 u 12 64 7 0.255 5281.60 2450.82

This shows that the system clock is out by 5.2816 seconds (slow) compared with the NTP master clock signal from central. What is more worrying is that the ntp daemon is only seeing a ntp packages (reach=7 - 377 is good, 0 is bad).

I hope Timekeeping in VMware Virtual Machines will give me some clues.