[ntp:questions] Re: Known asymmetry (DSL) compensation

[Richard B. Gilbert]

John Pettitt wrote:

>
> I know it's not possible to know or compensate for all asymmetry in 
> the packets between two machines.  However where there is a fixed, 
> known, asymmetry - like on a DSL line shouldn't it be possible to 
> calculate a fudge factor?
>
> Example:
>
> I have two FreeBSD boxes connected to the same GPS 18 LVC - they keep 
> time within 35 microseconds of each other but they are off with 
> respect to the rest of the net by about a 1200 microseconds.    The 
> connection from my local LAN to the Internet is a DSL line that runs 
> at 6000kbps downstream and 608 kbps upstream.   The calculated 
> difference is packet transmit time between upstream and downstream is 
> almost 900 microseconds.  I'd like to be able to bias out that known 
> asymmetry.
>
> If somebody who understands ntp internals  is willing to answer my 
> questions (off list) I'm up for taking a shot and implementing this in 
> a way that allows a bias to be specified and applied on a per net 
> basis - anybody up for it?
>
> John

I have seen the same sort of thing but, at least in my case I doubt that 
it is due to network asymmetry on my end.  I see consistant systematic 
offsets on a couple of the better servers I use.  One is usually +1.5 
milliseconds and another is usually -2 milliseconds with respect to 
GPS.  There are periods of the day when a graph shows this clearly; at 
other times there is sufficient noise or "jitter" that you would have to 
"fit" or "smooth" the curve to see it.

If a facility is added to compensate for these errors, it should be on a 
server by server basis since different servers can clearly show bias in 
different directions at the same time.