In this video I explain how a PTP slave synchronizes its clock with a PTP master through the exchange of four messages: the synchronization message, the follow-up message, the delay-request message, and the delay-response message.
There are a lot of videos on RU-vid about how the synchronization is done, but none of them actually explain exactly how its done. Yours does. Thank you for making such a nice video !!
It is easy to understand how propagation delay is compensated, I wasn't able to understand how delay inside the switch is compensated previously, but you explained very well. Thanks for making this video!
PTP cannot detect asymmetries. However, if the asymmetries are known, then PTP as defined in IEEE 1588 v2 can be statically configured so that clocks take them into account. This is called "asymmetry correction" in the standard.
Great video! The explanation was very easy to understand and the animations are very well done. The last step, delay_request and delay_response, assumes a symmetrical network topology. How can a PTP system detect or measure asymmetry, e.g. where master to slave takes 2 seconds but slave to master takes 3 seconds?
Hey David, do you know the significance of announce messages? What happens when you have 2 clocks connected and have different announce intervals (one set to 2(4 seconds) and one set to 4(16 seconds)? Great tutorials by the way! :-)
What is the rate of the messages (the bandwidth) between the Master and Slave ? and how does the message looks like, how long is each message ? is it possible to obtain the precision using RF link instead of fiber ? I would like to ask more questions if it is possible to contact you in someway it would be great. Thank you very much for the great videos
Same question asked by George in previous comment: "I understand why there needs to be compensation for the processing/encapsulating delay in the master's first message, but why isn't there a similar one for the slave's delay request time stamp? Wouldn't there be a delay between reading its own clock and transmitting as well?"
The Delay Request message does *not* contain a timestamp in the message. The slave reads the timestamp from hardware after the Delay Request message is sent.
because only the slave needs this timestamp, and master doesn't need it. Slave can read it's hardware timestamp after it sends out the Delay Request message.
Great Video for sure ! Question: why the Master could not embed the directly its transmission timestamps inside the first Sync message ? the Slave being capable to embed the transmission timestamp directly inside the delay_request message , I'm wondering why the Master would not be able to act similarly ? ( I would suspect here that add the timestamps inside the message would be done right before being sent on the wire, by the Network Interface itself) , thanks for your response !
How does it work if propagation is not the same in both directions? So when delay in one direction is not round-trip/2 ? That is not unrealistic - even in direct P2P links (though would be smaller than in NTP sync of a workstation over some dial up link or a antique broad band where downstream was via cable tv signal but upstream via dial up phone. This did happen in the early years where older cable networks where not return capable
That is why the timestamp of the Sync (1st) message is sent in the Follow Up (2nd) message. This gives the master time to fetch the recorded timestamp from hardware without any delay being incurred by the Sync message.
Thanks. it was very clear and nice step by step explanation. just one doubt: in last step why the calculated delay for propagation is divide by 2. it should be done in NTP because the time values that used to calculate the propagation delay include one way go and one way back. But here the time values to calculate delay just include one way propagation! Thanks in advance if you reply on this.
The slave time already contains the propagation delay from the follow up message (Master->Slave). Therefore, the Master->Slave propagation time is not required anymore.
I understand why there needs to be compensation for the processing/encapsulating delay in the master's first message, but why isn't there a similar one for the slave's delay request time stamp? Wouldn't there be a delay between reading its own clock and transmitting as well?
Hi George, the difference between the first master message (SYNC) and the first slave message (DELAY_REQ) is that the timestamp inside the message of DELAY_REQ isn't that important. The important value in this case is the time the packet left the port, which can be determined with high precision right away. I hope this helps.
Hi David, i work for CLARO ARGENTINA, having some doubts about implementing G8275.2 Telecom Profile; we found 10000 nano seconds in a TIME ERROR testing and ITU says the Maximun is 1500 ns; we have investigated on our packet network (QoS, link capacity) and no found any evidence of the problem. COuld you help me troubleshooting it? Regards
In the final delay request sent by the slave clock, you have chosen 208s, an even number. What if you had chosen 209s, yielding a difference of 3 seconds instead of 4? This would give us 1.5 second transmission delay - what happens then?
If the slave had sent its delay request at 209 seconds then the master would have received it at (according to its clock) a timestamp of 213. When the slave receives 213 in the delay response, it would subtract 209 from 213 = 4. So the propagation delay is still 4/2 = 2 seconds. However, if the network had a propagation of only 1.5 seconds instead of 2, the calculations are still exactly the same - you would simply add 1.5 seconds to the slave's clock.
