[clug] October Canberra Linux Users Group meeting

[Alex Satrapa]

On 22/10/2009, at 15:19 , Rod Peters wrote:

> The hosts get their time from an IPCop ntp server.  Seems to me that  
> the same
> would work for VM, but might require a port opened if NAT is used  
> between
> host & VM.  Could then counter drift by more frequent  
> synchronisations and
> the guests should not touch the host's clock.

The problem with VMs is not so much the drift while they're running,  
so much as what happens when you "suspend" a virtual machine (as  
opposed to telling the guest OS to suspend-to-disk or go to sleep),  
then "resume" that machine some days later. In this case, the guest OS  
has not received any indication that the hardware is going to sleep,  
nor has it received any indication that the hardware has been woken up.

One moment it's 11:00 on October 15 2008, the next it's being told by  
NTP that the date is actually 14:56 on July 26 2009. There's not much  
a system can to do cope with this kind of behaviour, and it's one  
reason I'm leaning towards the "crash cleanly and reboot from scratch"  
approach.

Alex