Comment by MichaelRenor | Hacker News Reader

Comment by MichaelRenor | Hacker News Reader

I disagree with the author. Before you go upgrading into ed25519, beware that the NSA/NIST is moving away from elliptical curve cryptography because it's very vulnerable to cracking with quantum attacks[0].

"So let me spell this out: despite the fact that quantum computers seem to be a long ways off and reasonable quantum-resistant replacement algorithms are nowhere to be seen, NSA decided to make this announcement publicly and not quietly behind the scenes. Weirder still, if you haven’t yet upgraded to Suite B, you are now being urged not to. In practice, that means some firms will stay with algorithms like RSA rather than transitioning to ECC at all. And RSA is also vulnerable to quantum attacks."

Stick with the battle tested RSA keys, which are susceptible but not as much as ECC crypto. 4097 or even better 8192-bit lengths.

There's no perceptible user benefits to using ed25519 and it's not even supported everywhere. Also you won't have to rotate all of your keys when workable quantum computers start crackin' everything.

[0] https://blog.cryptographyengineering.com/2015/10/22/a-riddle...

Both RSA and Elliptic Curve will fall to practical quantum computing, so the idea that you should use (weaker, slower) RSA today instead of curves because of QC is... dubious.

If you see quantum computing as a practical threat (because: you're encrypting and storing static data that needs to resist cryptanalysis for 20-50 years), you need to use a post-quantum cryptosystem. Unfortunately: nobody knows which pq systems will truly hold up to practical quantum computing, and it's possible --- even likely --- that implementations of pq schemes have dumb errors that nobody has thought to check for yet.

That's why Google's first foray into deploying pq crypto feeds both a curve computation and a Ring-LWE PQ computation into its KDF, so that unknown bugs in the RLWE exchange can't destroy the security of TLS. That's how you would seriously account for quantum computing in SSH.

The recommendations in this post are solid.

There is one argument to be made in favor of large key RSA when it comes to quantum computers: ECC may fall a bit earlier. There can be a situation where it's feasible to produce a quantum computer large enough to tackle a 255 bit ECC key, but not a 4096 RSA key. I have heard from (a few) cryptographers that this is a reason to stick with "old school" DH/RSA crypto.

I discussed this with various people involved in the postquantum debate and the general feedback I got was that this is likely not a big issue, because once quantum computers can be scaled it'll probably not be that hard to scale them up to RSA-breaking size. I think DJB once said something like "this will buy you a year" to me.

"battle tested" and "resistant to quantum computing" are complete opposites

Orthogonal?

Yeah.

Well, at the moment, they are "opposite" as well.

No battle tested crypto schemes are post-quantum safe, and no post-quantum crypto schemes are battle tested.

IAD recommends not upgrading to ECC in case this would incur significant costs, only to have to upgrade again once they settle on quantum resistant algos. So not because of crypto reasons, not because ECC is more vulnerable to quantum attacks relative to RSA, but because of practical operational and economic reasons. They think they will have quantum algorithms soon enough so that RSA >=3072 is OK in the meantime. At least that's my understanding of what they're saying here: https://www.iad.gov/iad/programs/iad-initiatives/cnsa-suite....

Are we not questioning the NSA's motives here given recent events?

"Don't use elliptical curve crypto because it's vulnerable to quantum attacks... keep using the stuff that is vulnerable to our in-house attacks"

Reply on news.ycombinator.com