Re: [Ntp] ntpv5 requirements

The reason that scalability and performance have traditionally been listed 
last in NTS documents is less that they are in any way secondary -- and 
more that they follow a pattern of "...and it needs to do all of the above 
in such a way that it retains scalability and performance as far as 
possible".
(And perhaps a bit of them being quantitative goals rather than 
absolute/qualitative; performance is gonna get worse with crypto rather 
than without, the goal is to keep it reasonable/best possible -- whereas 
e.g. amplification can be cleanly goal-stated as zero.)

Besten Gruß / Kind regards,
Kristof Teichel

__________________________________________

Dr.-Ing. Kurt Kristof Teichel
Physikalisch-Technische Bundesanstalt (PTB) 
Arbeitsgruppe 4.42 "Zeitübertragung"
Bundesallee 100
38116 Braunschweig (Germany)
Tel.:        +49 (531) 592-4471
E-Mail:   kristof.teichel@ptb.de
__________________________________________

Von:    "Harlan Stenn" <stenn@nwtime.org>
An:     ntp@ietf.org
Datum:  13.02.2023 23:29
Betreff:        Re: [Ntp] ntpv5 requirements
Gesendet von:   "ntp" <ntp-bounces@ietf.org>

On 2/13/2023 8:06 AM, Miroslav Lichvar wrote:
> On Thu, Feb 09, 2023 at 05:18:20PM +0000, Doug Arnold wrote:
>> For example: Judah Levine at NIST recently told me that he cannot 
implement NTS with his current server resources and the number of clients 
NIST supports.  However, when I told him about TESLA he thought a scheme 
based on that would be doable, as long as the keys didn?t have to change 
too often.
> 
> That is interesting as NTS was specifically designed to scale well to
> very large numbers of clients.

I don't recall performance in NTS as being a primary goal of the design.

Sure, it was listed as *a* goal, but the primary goals were around 
"security".

> Is their concern about decryption
> and encryption of NTS-protected NTP packets, or rather TLS in NTS-KE?
 >
> In 2016 they reported they had about 200k requests per second across
> all their servers [1]. Even if it was 100x more today and all clients
> were using NTS, that could still be handled by a dozen of servers with
> multi-core CPUs and AES acceleration. In my tests I get about 200k/s
> per core.

 From what I've heard, NTS key operations take 5-10x the amount of 
compute power beyond what NTP needs.

> NTS-KE traffic is more difficult to predict as it depends on the
> client implementations. I would be curious to see what NTS-NTP to
> NTS-KE request ratio do the well-known NTS providers like Cloudflare
> and Netnod have.
> 
> [1] https://nvlpubs.nist.gov/nistpubs/jres/121/jres.121.003.pdf
> 

-- 
Harlan Stenn <stenn@nwtime.org>
http://networktimefoundation.org - be a member!

_______________________________________________
ntp mailing list
ntp@ietf.org
https://www.ietf.org/mailman/listinfo/ntp