[ntpwg] REQUEST FOR TEXT - Protocol Doc Changes

[STUART VENTERS]

With regards to where to put the timetag.

The simplest plan is certainly to put it at the start of the packet like 1588 etal.
  I think I still side with simple,
  but Dave's comment about how things happen in the kernel made me think of the following:

A S/W implementation doesn't have a single place to put the timetag.
   For Transmit, the tag might be just a bit before the first bit.
   For Receive, it might be a bit after the last bit.
   Assuming the packet times are the same in both directions,
     these differences cancel out just like link delay.

I thought this was a bug due to what was possible from the constraints of NIC's and kernels,
   but it may actually be a feature if the server and client are connected to a switch via different speed links.

Assuming the following:
	1) a switch forwards after receiving a whole packet,
	2) the server is connected with a fast link
	3) the client is connected with a slow link

Then with the timetag at the start,
   the client to server delay is processing time plus a few bit times plus a packet time on the slow link,
   the reverse direction is similar except with a packet time on the fast link.

This causes a violation of the symmetric delay requirement and gives the client a time offset.
With the timetag like the S/W, it looks like the offset goes away.

Note that changing the number of header bytes is another way to change the effective link speeds.

Hopefully, I'm not drawing my time lines correctly?

Cheers,

Stuart