Again Time Drift Problems on 64bit Windows Guests

Hi everybody,

after some Problems last year with time drifts on 64bit Windows Guests and the use of
qemu HPET Emulation, which fixed my issues, the problem started again.

My situation:

Under some load (CPU or I/O), the internal clock on the Windows Guest starts to lag behind. On heavier load as much as up to 10+ seconds per minute.
As these machines are on a Windows Domain functioning as Terminal Servers with access to some File and SQL Servers this behaviour results in massive problems e.g. with authenticating and invalid Kerberos Tokens.

With PVE 1.5 the Problem dissapeared in my opinion because of the HPET Emulation of QEMU. After a upgrade to the latest 1.7 Version of Proxmox (with a newer Version of QEMU obviously without HPET) the clock problems started to appear again.

My (not very satisfactory resolution) is to sync the Windows Guest with the Domain Controller every 2 minutes (which is synced with an external NTP time source)

Has any reliable way to avoid the time drift been found?

Thanks in advance

With QEMU 0.13+ HPET Emulation was removed (at least the qemu change logs indicate this) and whenever i typed "info hpet" into the qemu monitor it showed me "HPET Emulation enabled/disabled" with QEMU versions proir to 0.13.

Now with PVE 1.7 and QEMU 0.13 the "info hpet" command doesn't display this information anymore.

Okay thanks for the info!

But my problem sadly remains the same.
I now activated HPET as clocksource on the host, but that didn't change anything.
Still syncing with the Domain Controller every 2 minutes....
The only difference between my former setup and the setup now are virtio block and network drivers within the Windows machine....

I noticed a ridiculously growing skew at the time my robocopy script backups to another server via network. I presume a skew about 800 seconds in i little bit over half an hour is not that nice

Is virtio known to have such effects? Before the update to 1.7 I used e1000 NICs.

Should I revert back to e1000 or is it an option to use a qemu version lower than 0.13, e.g. 0.12.5?

Is it maybe the physical platform? (I'm using an Intel Modular Server)

Thanks in advance

The problem so far seems to be resolved if you don't use VirtIO network drivers within the guest.
At least in my case....

After some experimenting I removed virtio nic drivers, gave my guests a e1000 nic and installed the latest e1000 drivers from the Intel Website.
Now the clock runs perfectly....

But that kind of problem with virtio drivers was new to me.....