[Pool] Virtualization

lst_hoe02 at 79365-rhs.de lst_hoe02 at 79365-rhs.de
Tue Jul 21 14:24:07 UTC 2015

Zitat von Ask Bjørn Hansen <ask at ntppool.org>:

>> On Jul 20, 2015, at 12:29 AM, lst_hoe02 at 79365-rhs.de wrote:
>> The problem with virtualization is not that it doesn't work at all,  
>> but you can never be sure. If you have dedicated hardware you are  
>> able to monopolize CPU to do very accurate counting/timing  
>> operations, […]
> But for most actually deployed servers isn’t that almost as true  
> when you’re not virtualized? Modern kernels and CPUs do all sorts of  
> tricks to save power and divide resources. Most servers aren’t  
> configured to run at full speed just in case it makes the time  
> keeping better.
> Ask

That's the problem, it is already difficult to get a precise time  
tick/measurement on bare hardware, but with a Hypervisor in between  
which is trying to isolate the hardware from the VMs running, it get  
worse in some cases because the VMs are not aware of the decisions a  
Hypervisor does to manage hardware/counter access. This is *never* a  
problem with "standard" use cases but it might get a problem with many  
VMs all trying to get reliable time ticks for example.

There are some internals here:

But as already said i doubt this is a *real* problem for pool members,  
maybe Stratum 1 should care but that is more of a question targeting  
the NTP developers i guess.



More information about the pool mailing list