View Issue Details
|ID||Project||Category||View Status||Date Submitted||Last Update|
|0005581||CentOS-6||time||public||2012-03-08 16:40||2014-08-29 13:54|
|Summary||0005581: Constant clock drift under Hyper-V Host|
|Description||I'm running 6.2 under a Hyper-V host.|
I'm seeing a pretty much constant increase in time of approx 0.0625 secs a minute, or 90 seconds per day - i.e. the centos clock is running too fast.
I'm using version 3.2 of the MS linux integration components with time syncronization enabled.
As I understand it, the Centos 6 kernel should no longer require settings to be added to the bootloader for it to work in a virtualized environment.
cat /sys/devices/system/clocksource/clocksource0/current_clocksource shows hyperv_clocksource
|Steps To Reproduce||Create a Hyper-V VM with 1 virtual CPU.|
Install Centos 6.2 and MS linux IC 3.2 on the guest.
Now run inside the VM ntpdate -q 0.centos.pool.ntp.org
Wait a while, run it again. You'll see the clock running faster than it should
As soon as the machine starts up, the time appears to be correct.
However, if I run ntpdate -q 0.centos.pool.ntp.org every minute the difference gets greater every time I run it.
It is 100% reproduceable, and appears to be at a constant rate.
The clock on the host remains the correct time.
|Tags||No tags attached.|
|Kernel version is 2.6.32-220.4.2.el6.x86_64|
I see the same behaviour in Cent OS 6.5 and Cent OS 7 (integrated hyper-v drivers) on hyper-v host 2008 R2.
Clock gains similar to yours 10s of seconds per day and always fast.
I use NTP to mitigate this but sometimes the jitter is too much for it and it loses sync.
Also I note that saving and restoring (we do this for backup as no VSS for Linux on 2008R2) occasionally does not set the time on restore and the clock does not recover from this.
I am not sure how hyperv_clocksource works but presumable there is no actual communication of time of day and it is more of a tick provider.
ie it never compares the time of the host and guest.