[ntp:questions] running NTP as server only
maarten at kittensandcats.net
Thu Aug 19 11:58:05 UTC 2010
"Rob" <nomail at example.com> wrote in message
news:slrni6q3i8.ie0.nomail at xs8.xs4all.nl...
> unruh <unruh at wormhole.physics.ubc.ca> wrote:
>> On 2010-08-19, Rob <nomail at example.com> wrote:
>>> unruh <unruh at wormhole.physics.ubc.ca> wrote:
>>>> On 2010-08-17, folkert <folkert at vanheusden.com> wrote:
>>>>> Is it possible to run the NTP daemon only as a server and not as a
>>>>> local-clock maintainer?
>>>>> Reason: I have a virtual machine which gets its time via the vmware
>>>>> tooling from the hardware server it is running on. Now this virtual
>>>>> machine needs to distribute the time to clients.
>>>> Aarrgdaagh. Why would you have a virtual machine, with its
>>>> remarkably unreliabl e clock serve its time to others?
>>> Some companies are virtualizing all their hardware.
>>> E.g. this is happening where I work as well. All the servers have
>>> been replaced by a number of Vmware ESX machines.
>>> So there is no physical hardware machine left to run as the ntp
>> uh, to quote Landauer, all information is physical. All virtual
>> machines MUST also run on physical machines.
> But that does not mean you can run NTP on them.
> E.g. on VMware ESX, you cannot do this.
> (there is an NTP running on the console session, but that is just a
> virtual machine running a Linux variant, it is not running on the
> physical machine either)
And not so long ago, somebody quoted a more recent whitepaper here
that said running NTP in the virtual machines was now working much
better and in fact recommended.
In situations where the host has enough cores that they can be
allocated more persistently to running VMs, I can even see this work.
The problem used to be, mostly, that interrupts didn't (always) arrive
on time. Yet another problem solved by throwing hardware, to wit
interrupt lines, at it.
More information about the questions