Thank you Dr. Kleppmann, you really make me get more interested in learning Distributed Systems. I've already bought your published book to study and learn more!
6:30 the total network delay can be found by looking at time stamps but this assumes network symmetry We can estimate what t4 was supposed to be and compare with actual. The estimate is t3+(delta/2). Difference between actual t4 and expected is the clock skew
@Martin, thanks for sharing. I'm new to this subject and trying to understand how to convert Time Error (drift in seconds per day) to precision in positioning. There are claims suggesting TE at 10^(-18) provides cm precision. If we multiply 10^(-18)*86400 (sec/day)*speed of light, the precision should sub cm. I would be grateful if you help with a comment or reference. Many thanks.
I am thinking resetting the client clock happens if the skew is larger instead of slewing as in smaller skew case is because synchronizing the clocks would take a long time otherwise.
In NTP why can you do (t4 - t1) - (t3 - t2) ? First two terms are from the client and second two terms are from the server, and if the two clocks have different rates than aren't you working with different units? Or is the assumption that the rates are 'close enough', i.e. corrections to the corrections are second-order?
I can't understand the slew chart at 10:06. If the clock ended being synchronized at ~200th mark (miilisecond?) itself, why was the quartz still ticking slowly until ~650 ms?