Re: [Ntp] Should NTPv5 have QUIC bindings?

[Hal Murray <[e-mail]>]

[e-mail] said:
>     > What problem(s) do you think it will solve?
> Nothing definitive which is why I asked.  The IETF is moving to QUIC for many
> things, so I'm curious. 

I think you neet to talk to a QUIC wizard.

I'm assuming the idea is to replace NTS with QUIC.

One of the goals of NTS was to avoid per-connection storage on the server.  It 
uses TLS to get cookies that hold the connection info.  I don't see how to do 
that with QUIC.

As I understand it, if you are interested in performance, there are 2 types of 
crypto.  RSA is slow.  AES if fast.  AES requires shaared keys.  TLS uses RSA 
to setup a connection that holds the shared key.

I don't know enough about QUIC to count packets.  It looks to me like QUIC 
would either have to retain connections on the server or go through the 
connection setup for each NTP exchange.

It's worth considering retaining connections.  Again, I don't know any QUIC 
details, but memory is cheap these days.  It's worth considering keeping a 
connection for each client.  That might not work for something like the NIST 
servers but it could work for almost everything else.

----------

I saw mention of QUIC not using the IP Address for a connection key.  That 
sounds like an invitation for easy tracking.  If we take non-tracking as a 
serious goal, that could be enough to knock QUIC out of consideration.


-- 
These are my opinions.  I hate spam.