Re: [Ntp] Circular dependencies

[Warner Losh <[e-mail]>]

On Sat, Jan 16, 2021, 7:57 PM Seaman, Robert Lewis - (rseaman) <
[e-mail]> wrote:

> Hi Warner,
>
>
>
>    - Sure makes cold spares hard to implement.
>
>
>
> You’ve often made this point. Aren’t there other services that provide
> similar challenges? Starting with DNS for that matter. Some engineering
> requirements are indeed hard to implement. Coherent precise timekeeping may
> simply be one of those.
>

Yes. There are many challenges. Making precise time a requirement for DNS
in order to get the time is not a circular dependency that should be
undertaken lightly. At least with a GPS receiver you'll know the time
within a minute or two for the rest of this century even without the
almanac. Absent a GPS receiver, it can be hard to know time well enough to
trust for all things that follow. If you can control time, and time is the
basis of trust... look out.

I am just saying that the quick hacks that were suggested won't work. Any
extended downtime now needs a trusted way to bootstrap even out of date
systems that have been off and missed important updates and revocations.
Any design under consideration should use that test case to proof the
design. I'd like it to be considered and have a protocol that's proven
secure rather than everyone rolling their own. We shouldn't wave our hands
and say it is hard, but smart people will figure it out. Any proposed
system has to cope, or it simply won't work when deployed.

Things have to work both in the steady state case, and in the large step
case. Usually at least one of these is hard.

It trips up lots of designs which is why I bring it up in a lot of
contexts. :)

Warner

Stay safe,
>
>
>
> Rob
>
>
>