Resuming the great cleanup

Achim Gratz Stromeko at
Tue May 29 20:18:39 UTC 2018

Hal Murray via devel writes:
> devel at said:
>> Provided you somehow route the packets from different networks to different
>> (virtual) interfaces, you could measure and eventually compensate the
>> asymmetry associated with that particular network provided you have some
>> trusted time server on each side of the asymmetry. 
> There is a tangle between interface and routing info.  An interface can be 
> asymmetric.  But the asymmetry can also be one or more hops away.  You might 
> be able to catch that case based on IP address.

That's what I'd do by routing said packets to different virtual interfaces.

> This could turn into a bottomless pit.  There could be a router on the far 
> end of an asymmetric link that routes to many CIDR blocks.

Yes.  The general case is _really_ complicated and ugly and would need
some different protocol than NTP to solve properly.  It's doable
approximately if you have enough sources of "good enough" absolute time
that feed NTP servers on many network nodes (a properly set up and
maintained GPS-based stratum 1 chimer is "good enough" in that it should
have a smaller error than the inevitable noise from other sources).  I
think there's one paper that calls this "network tomography", based on
the deconvolution of sorts you'll have to do to figure out the delays.

For instance, I see much larger temporal variability of these asymmetric
delays to the PTB servers for some time now.  It seems the packets are
now going through some software defined routing mess from Telia for me
(which is also almost twice as long as the shortest network path).  I
know it's the routing since I can see this from five GPS-backed servers
which I can correlate to each other.  That also shows me that there is
some sort of round-robin load-balancing going on for the three public
facing servers.  Last but not least I can see when I'm using up the
bandwidth fully on my connection.

+<[Q+ Matrix-12 WAVE#46+305 Neuron microQkb Andromeda XTk Blofeld]>+

Factory and User Sound Singles for Waldorf Blofeld:

More information about the devel mailing list