[Ntp] Antw: [EXT] Re: NTPv5 client cookie size

>>> Miroslav Lichvar <mlichvar@redhat.com> schrieb am 05.11.2022 um 10:01 in
Nachricht <Y2YmgLTrMXKOEIGl@localhost>:
> On Fri, Nov 04, 2022 at 12:03:38PM +0100, David Venhoek wrote:
>> Dear all,
>> 
>> After having implemented the NTPv5 spec, I feel that the current size
>> for the client cookie it gives is on the small side. Although 64 bits
>> is probably still hard to guess for an out‑of path attacker, I think
>> we should seriously consider increasing its size given that 64 bits of
>> security is usually regarded as insufficient.

I think Andrew S. Tanenbaum explained in one of his books that (at that time
illusional) if a computer would count at a rate of 1GHz, it would take almost
213504 days (about 585 years) until the timer is exhaused. So I wonder what
kind of problem you are trying to fix. I also think (maybe it was a book of
Stephen Hawkins) that the number of atomes in thge universe is estimated to be
less than 2^128.

Regards,
Ulrich

> 
> Against what attacks would the increased cookie length help?
> 
> The cookie is not a key that an attacker could be cracking offline.
> It's supposed to protect against off‑path attacks. The packets need to
> be received by the client. A typical client running on modern HW
> cannot handle more than million responses per second. That's about
> 300000 years on average to guess a single 64‑bit value.
> 
> Even if we assume an infitely fast CPU and network speed of 400Gb/s,
> it's still about 500 years of guessing.
> 
> If we consider a client using random source port, those numbers need
> to be multiplied by at least 10000 or more depending on the OS.
> 
> ‑‑ 
> Miroslav Lichvar
> 
> _______________________________________________
> ntp mailing list
> ntp@ietf.org 
> https://www.ietf.org/mailman/listinfo/ntp 