virtualbox.org

Hi all.

I was looking into which services could be disabled on my Windows VM, and realized that the Windows Time service was probably of no use with guest additions installed.
I tried to find some documentation to support this, but couldn't find any.

Common sense says running an NTP client concurrently with the guest additions is a very bad idea, though I've never had a problem with it.
All of my Windows VMs run happily with guest additions and the Windows Time service enabled.

So what's your take on this, people?
Do you just leave Windows Time enabled, because it won't harm anyone? (or will it?)
Do you religiously disable it on every new Windows host?
Do you think the guest additions should take the extra step to disable it on install?
Or do you just not care a bit about this?

I don't worry about it. But I guess if you feel the need go for it.

Well, I've been to the other side of the street to see what they have to say about this.
According VMware, on their platform, of the VM tools and the guest NTP client, each one has its advantages and pitfalls, and you should use either one, but not both.
They have quite some stuff on the subject, and while it may not apply to VirtualBox at all, it comforts me that clock synchronization is worth digging into.

Then I did some experimentation with VirtualBox, moving the guest clock ahead or past the host clock, with the guest additions installed and the Windows Time service stopped.
What I did see is that once the guest time gets too far from the host clock, the guest additions indeed kick in, and the result is... surprising.
The length of the 'virtual second' is adjusted, either shortened or lengthened, in order for the guest clock to catch back with the real clock.
The problem is that this adjustment is not very sound. If the guest time is, say, 5 min ahead of the host time, the virtual second length gets longer, longer and longer, to the point where the two clocks will cross, and the guest time is now late! Then only, the virtual second will start getting shorter, again, again and again, and the guest clock will be (a few seconds) ahead of the host again!
The guest clock ends up wobbling around the host clock and seems to never really stabilize.

In the end, the whole process does not look very reliable.
While one might never get there if no one is tinkering with the guest clock in the first place, but there's definitely room for improvement in the guest additions' clock management.

So there's no final conclusion here.
I'll try to run some Windows guests with the Windows Time service disabled, and see what happens.

You know that it works fairly the same in the real world on metal right? If a user changes their time and they are on a domain, it can cause all kinds of problems and the NTP service does not adjust that often either.

I have only encountered one problem with the guest additions timesync. If it corrects the guest time it freaks out the Linux mail server (dovecot) when it changes it to the negative (guest went to far forward and was corrected down). In that case you must disable the timesync and just hope for the best.

What is the reason for all of this (just so I know what you are really trying to do)?