[ntp:questions] Creating a multi-minute offset

Terje Mathisen "terje.mathisen at tmsw.no" at ntp.org
Fri Feb 24 10:35:18 UTC 2012


A C wrote:
> On 2/22/2012 23:56, David Woolley wrote:
>> A C wrote:
>>> Ok, so here's an odd question but I had an amusing thought so I want
>>> to know:
>>>
>>> Is it possible to force a client to synchronize with network servers
>>> but also maintain a time offset of several minutes from those servers?
>>
>> If this isn't just a special case of timezones, you would need to modify
>> the OS interface layer of the code to shift the time when reading and
>> writing the system clock, so that the core of ntpd sees the correct time.
>
> Yeah, it's not a matter of time zone, it's trying to force an absolute
> time lag on one machine as part of a simulation. The idea was that the
> clocks should run at the same speed but one of them needs to be shifted
> from the other.
>
> It was a thought but obviously it doesn't work.

This is perfectly possible, but it does require a small source code patch:

The proper way to do it is to modify all the fudge code so that relevant 
parts can be deployed on network servers.

I am thinking particularly on the ability to measure (using a temporary 
local refclock) any constant time bias on a non-symmetric uplink, like 
many/most ADSL links, and then fudge that out.

Such a facility would also be able to handle your desire for a simulated 
time frame.

(In fact, it could also be used to easily test how a particular system 
would handle a previous or upcoming leapsecond, by fudging the reference 
enough months backwards/forwards.)

This is not a trivial job however.

Much easier is to grab the source code and make a local patch where you 
simply add/subtract a constant offset to all timestamps received from 
network servers.

This could be done in the ntp packet parsing code, with a two-line 
patch. :-)

If you do so, _please_ make absolutely sure that you never provide time 
service to anyone outside your test network!

Terje
-- 
- <Terje.Mathisen at tmsw.no>
"almost all programming can be viewed as an exercise in caching"



More information about the questions mailing list