[ntp:questions] determining time difference between two not-directly-connected computers

Danny Mayer mayer at ntp.isc.org
Sun Jan 7 05:30:59 UTC 2007


matthew.garman at gmail.com wrote:
> I've got two (x86_64 Linux) machines between which I need to
> determine the relative time difference.  The problem is, the
> machines are not directly connected and only one is connected to the
> Internet.
> 
> The topology looks like this:
> 
> box_1 <--VPN--> box_common <--LAN--> box_2
> 
> It's not so important that box_1 and box_2 are synced, but that
> their time difference is accurately measured.
> 
> So, box_1 and box_common sync via NTP to (different) public stratum
> 2 NTP servers.  box_2 syncs to box_common.
> 
> I wrote a simple program to try an determine the time difference
> between box_1 and box_2.  The program works like this:
> 
> box_1 and box_2 run the program in "server" mode; when queried, the
> server program calls gettimeofday() and returns the timeval result
> via UDP.
> 
> box_common runs the program in "client" mode.  client mode does the
> following:
> 
>     - record the current time via gettimeofday() as t1
>     - send a UDP packet to the remote server
>     - wait for the UDP response (i.e. the remote machine's actual
>       time)
>     - record the time of the UDP response as t2
>     - define "remote calculated time" as the midpoint between t1 and
>       t2
>     - calculate the difference between "remote calculated time" and
>       the actual remote time as provided by the UDP response; call
>       this t_d
> 
> Once I've done the above for box_1 and box_2, I subtract t_d1 and
> t_d2.
> 
> I'm just looking for some feedback on my approach: is it valid (i.e.
> am I overlooking something?), are there better approaches, etc?
> 

What do you think that NTP does? and why don't you think that after 20+
years of engineering that it doesn't do a much better job?

Danny



More information about the questions mailing list