4 ms·

It's basically impossible to measure the one-way latency, without external information in the form of clock synchronisation done externally. See http://twisted

It's basically impossible to measure the one-way latency, without external information in the form of clock synchronisation done externally. See http://twistedoakstudios.com/blog/Post2353_when-one-way-latency-doesnt-matter http://twistedoakstudios.com/blog/Post2353_when-one-way-late... This fact is also the basis for special relativity (different observers may choose different simultaneity conventions - for example Einstein synchronisation)

bentcorner 6y ago
You should be able to tell when one side of the link degrades though, correct? You'd have known time deltas for the link each way (whether its accurate doesn't really matter) and when the round trip time changes you should be able to tell which side that occurred on. (I'm interested in this problem because my cable internet reliably (!) degrades during the workday)
- slaymaker1907 6y ago
  The problem is physics can't even distinguish whether light travels instantaneously in direction x, but at c/2 going in the opposite direction. There is no known way to synchronize clocks or come up with a clever apparatus to prove that light moves at the same speed in every direction. You need to make some strong assumptions in order to try and measure asymmetrical latency on a network. I believe the method you are proposing is to measure RTT when RTT differs depending on the starting node. The difficult problem is determining how much of your RTT time comes from node A to node B as opposed to node B to node A. I believe you need to at least have some sort of oracle on both sides which can synchronize time without using the network path you are trying to measure (I think you can do this if you have node C which both A and B have a fast connection to relative to each other, this implies that the delay between A and B is significantly longer than the speed of light delay).
datastoat 6y ago
To be precise, it's impossible to measure one-way latency between a pair of nodes without external information. But if you have a mesh of nodes, the story is different [1]. If you have N nodes and hence N unknown clock offsets, and if you have ping data from N(N-1) pairs, you can do better. [1] https://www.usenix.org/conference/nsdi18/presentation/geng https://www.usenix.org/conference/nsdi18/presentation/geng
- Ono-Sendai 6y ago
  Well I think we need to distinguish between practical latency measurements on earth, where it's mostly a matter of rejecting dodgy data / outliers etc.., and detecting systematic latency bias that is theoretically impossible to detect. If you could get around this problem just by using a network of N nodes, then you could do stuff like measuring an absolute reference frame, thereby violating special relativity.

bentcorner 6y ago

You should be able to tell when one side of the link degrades though, correct? You'd have known time deltas for the link each way (whether its accurate doesn't really matter) and when the round trip time changes you should be able to tell which side that occurred on. (I'm interested in this problem because my cable internet reliably (!) degrades during the workday)

slaymaker1907 6y ago
The problem is physics can't even distinguish whether light travels instantaneously in direction x, but at c/2 going in the opposite direction. There is no known way to synchronize clocks or come up with a clever apparatus to prove that light moves at the same speed in every direction. You need to make some strong assumptions in order to try and measure asymmetrical latency on a network. I believe the method you are proposing is to measure RTT when RTT differs depending on the starting node. The difficult problem is determining how much of your RTT time comes from node A to node B as opposed to node B to node A. I believe you need to at least have some sort of oracle on both sides which can synchronize time without using the network path you are trying to measure (I think you can do this if you have node C which both A and B have a fast connection to relative to each other, this implies that the delay between A and B is significantly longer than the speed of light delay).

datastoat 6y ago

To be precise, it's impossible to measure one-way latency between a pair of nodes without external information. But if you have a mesh of nodes, the story is different [1]. If you have N nodes and hence N unknown clock offsets, and if you have ping data from N(N-1) pairs, you can do better. [1] https://www.usenix.org/conference/nsdi18/presentation/geng https://www.usenix.org/conference/nsdi18/presentation/geng

Ono-Sendai 6y ago
Well I think we need to distinguish between practical latency measurements on earth, where it's mostly a matter of rejecting dodgy data / outliers etc.., and detecting systematic latency bias that is theoretically impossible to detect. If you could get around this problem just by using a network of N nodes, then you could do stuff like measuring an absolute reference frame, thereby violating special relativity.