[Cryptech Tech] goals / use cases

Randy Bush randy at psg.com
Sat Jan 24 12:19:01 UTC 2015


>> does it suffice to divide the use cases into 2-3 taxa?
>>   o key generation/store/sign/validate/... (dnssec, rpki, tor consensus)
>>   o low speed encryption (pgp, small vpn, ...)
>>   o encryption (tls, vpn, ...)
> 
> That's really just a way of saying "everything" though.

not exactly.  it was meant as a way of dividing the problem so we could
say "we can do X now, but not Y."

> I think we need some more concrete targets, for example how much RAM
> will be available, and will it be dynamically or statically allocated?
> How many keys can be stored?  How much flash will be available?

those are weapons, not targets.

> For example a single statically-allocated (at system build time) RSA
> key is very different from fifty dynamically-allocated { RSA, DSA,
> ECDSA, DH } keys in terms of memory usage and requirements.  How long
> should a 2Kbit private-key operation take?  A millisecond, a hundred
> ms, a second, ten seconds?  Without some sort of yardstick to use as a
> target, it's hard to tell whether we're winning/losing, or even on the
> right track...

now you're one level up, which is good.  but i think we answer these by
decomposing the needs of the use cases, which i tried to start dividing
according to my weak understanding of the scale of resource needs (and
basil chopped a bit more (too late in evening to look for his message)).

so, for giggles, let's say serious tls and circuit level crypto are a
later design, and the alpha target is dnssec signing (and validation?),
rpki, tor consensus, and, for giggles, pgp (at human rates).

folk such as rob, jakob, and linus (on family leave) should be able to
transform the dnssec, rpki, and tor consensus use cases into estimates
of the number of keys, rate of change, algorithms, need for speed to
validate, etc.

randy


More information about the Tech mailing list