Hacker News Reader

Comment by Xarodon | Hacker News Reader

Because they know the speed of light and the distance between nodes so they can account for the propagation delays of light due to distance.

They're not talking of sub millisecond latency in communications.

View on HN

No, we don't know the distance between nodes (although we could deduce it). But using timestamps, we can know the round-trip time.

See https://www.ohwr.org/project/white-rabbit/uploads/2b9d42b664... (page 9 and later for the principle).

If you want all the details, see https://ohwr.org/project/white-rabbit/uploads/6a357829064b9e...

When synchronizing two nodes A and B, where there is a persistent difference in the travel times A->B and B->A, how do you achieve synchronization when knowing A->B->A or B->A->B?

For a first approximation, you can assume A->B and B->A travel times are equal.

And because optical links are used, the asymmetry is mainly due to the wavelength difference which is known.

you can't. You can only assume that they are equal and attempt to make them as equal as possible. (the same issue arises when measuring the speed of light: it's actually not possible to distinguish if the speed of light is different in one direction to another, we only know accurately the average of each direction)

Delay symmetry is a critical assumption in any two-way time transfer process. White Rabbit goes to extreme lengths to maintain that property.

This includes mandating use of cables that share a single optical fiber, with specific wavelength pairs and fiber types so you can calibrate for unavoidable differences in propagation time.

More info on their wiki:

https://ohwr.org/projects/white-rabbit/wiki/SFP

What happens when the roundtrip time isn't consistent?

The roundtrip time is never consistent. Light travels with different speed in fiber depending on the temperature. This is why you calibrate every second.

Even better, the actual in situ delays are measured and compensated for, and it works independent of the physical connection (and through fiber/copper, switch layers, etc.).

And even better: they suggest the use of a single medium for both transmission and receiving (1000BASE-BX10) to minimize asymmetry.

https://ohwr.org/project/white-rabbit/wikis/SFP

Only over fiber. Copper SFPs are not deterministic enough to precisely synchronize networks.

Indeed. It's exactly the same (albeit on a different scale) as NTP synchronization, where you can frequently (ha!) reach a few ms accuracy over a hundred ms latency network.

Reply on news.ycombinator.com