[ILUG] Timekeeping in virtual machines

[Cillian Lyons]

Hi Niall,

On Thu, 2008-04-03 at 11:08 +0100, Niall O Broin wrote:
> Timekeeping in virtual machines has definite issues, whether the host  
> software be Xen or VMware (or presumably some of the others, though  
> my experience is mostly with Xen and VMware) and my specific question  
> now concerns VMware workstation, where I have a VM running SLES9. The  
> matter is discussed at http://www.vmware.com/pdf/vmware_timekeeping.pdf
> and what I currently have is
> 
> tools.syncTime = TRUE
> 
> in the vmx file, and in the kernel options line
> 
> clock=pmtr
> 
> I've also tried
> 
> clock=pit nosmp noapic nolapic
> 
> in the kernel options line but it was no different from clock=pmtr.
> 
> NTP client software is NOT running in the VM and now time tracks the  
> host reasonably, but jerkily, like this:
> 
> while true; do here=`date +%s`; there=`ssh host "date +%s"`; echo -n  
> " $[$there - $here]";sleep 1; done
> 
> 4 5 4 5 5 6 5 5 6 5 6 6 7 6 7 7 7 7 8 8 8 9 8 8 1 0 1 1 1 1 2 2 2 2 3  
> 2 2 3 4 3 3 4 3 4 4 4 4 5 5 5 5 5 6 5 6 6 7 6 7 7 7 7 7 8 7 8
> 
> For my purposes now it's OK, but I can certainly see situations where  
> it would not be. For instance:
> 
> while true; do then=`date +%s`; sleep 5;now=`date +%s`; echo -n " $ 
> [$now - $then]"; done
> 
> 5 5 13 5 5 5 5 5 5 5 5 5 5 5 13 5 5 6 5 5 5 5 5 5 5 13 5 5 5 5 6 5 5  
> 5 5 5 5 13 5 5 5 5 5 5 5 5 5 5 5 15 5 5 5 5 5 5 5 5 5 5 5 13 5 5 5 5  
> 5 5 5 5 5 5 5 14 5 5 5 5 5 5 5 5 5 5 5 13 5 5 5 5 5 5 5 5 5 5 5 13 5
> 
> Sometimes after sleeping for 5 seconds, you see a wallclock advance  
> of much more. I can definitely imagine scenarios where this would be  
> bad.
> 
> I have also tried the suggestion in "Guest Clock Synchronization With  
> Non-VMware Software" in the referenced document, but after 20 minutes  
> up, the guest was two
> minutes behind the host with ntpq -p showing
> 
>       remote           refid      st t when poll reach   delay    
> offset  jitter
> ======================================================================== 
> ======
> *tardis.local    84.16.251.78     3 u   13   64  377    0.061   
> 122898. 584.872
> 
> 
> 
> To be honest, I expect that "time tracks the host reasonably, but  
> jerkily" is as good as I'm going to get but if anybody has any other  
> suggestions, I'm all ears - or eyes.
> 

Time-keeping in Virtual Machines can be affected by a number of factors.
One such factor is clock interrupts. 

Did you come across VMware KB Article: 1420? 
http://kb.vmware.com/selfservice/viewContent.do?language=en_US&externalId=1420

2.6 kernels request interrupts at 1000Hz, while 2.4.x were at 100Hz.

If you are in a position to recompile with clock interrupts at 100Hz,
you may see better time tracking. 


Hope this helps,

Cillian