[Ntp] Alternative for interleave mode

[David Venhoek <[e-mail]>]

Dear All,

The current proposal for NTPv5 includes an interleaved mode. It's
current design is quite involved, with the server keeping state across
multiple exchanges, and a client potentially having to wait a further
polling interval before getting the final timestamp. As such, it makes
implementation a bit tricky.

I would like to at least have discussed the following alternative:
When a server receives an "high-precision" request, it first sends to
the client the actual response, and then immediately on getting the
transmission timestamp for that response, it sends a second response
with the higher-precision transmission timestamp (similar to PTP's 2
step mode). This would have the advantage that a client gets its high
precision timestamp immediately, without having to do multiple NTP
requests in short succession, and that the server would not have to
keep any long-term state.

I know this probably immediately triggers a gut reaction of: "But then
the NTP server can be used to amplify traffic in reflection denial of
service attacks". Yes, technically we indeed now get a greater than
unity amplification factor. However, the amplification factor is still
only 2, both in bytes and number of packets, which is well below other
widely deployed UDP-based protocols such as quic (and from what i can
tell, quite well accepted, the quic working group is working by the
guideline that one should never send more than 3x the packets/bytes
received before validating origin). Given the simplicity of the above
scheme compared to interleave mode, I am of the opinion this is a
tradeoff worth making, but I was wondering what other peoples view is?

Kind regards,
David Venhoek