[Cryptography] RNGs, Entropy, and Unguessability (really clock synchrony)

[Natanael]

Den ons 9 juni 2021 00:33Robert Wilson via cryptography <
cryptography at metzdowd.com> skrev:

> A few millennia ago I was working on "provable" security. (Orange book
> stuff.) Lots of people wanted to use inductive proofs: Start in a known
> good state, prove that all possible transitions maintain what you want,
> voila.
>
> We had to consider distributed systems. It was obvious to me that there
> is no place in the universe (per relativity) where one could know a
> well-defined state. What could it even mean?
>
> I had then been thinking of distributed as meaning something like
> network-connected machines. But as Kent correctly points out, even
> within what we call a CPU chip, at any instant (and what does even that
> mean?) any distinct points will have different notions of what's going
> on! What kinds of systems, if any, can have well defined states?
>
> Bob Wilson
>

This is what for example concensus systems and atomic databases are for,
along with related systems like append-only logs, logging filesystems, etc.

For cryptographic solutions, we got Zero-knowledge proofs and related forms
of proofs of correctness, paired with formal logic.

On the hardware level there's error correction, and schemes like using
[tmultiple processors which need to agree on a result (the NASA approach),
etc.

We do not attempt to make sure the complete system is perfectly well
defined in all components at all times - instead we select a carved out
subset which we try to ensure will be in a coherent state. We try to ensure
that whenever you poll the system you can get a valid response out.

>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://www.metzdowd.com/pipermail/cryptography/attachments/20210609/19ddc135/attachment.htm>