Today, all you need is a 32-byte ECDH private key, and a corresponding 32-byte public key. To talk to any of the billions of devices on the Internet securely, all you need is their 32-byte public key, and some means to verify that it is authentic (i.e. certificate authorities + certificate transparency). With this information, you can calculate a shared secret (or use a sealing API given only their public key, if you really want).
Total local storage: PKI overhead plus your own keys, under a few megabytes.
Doing PSK at scale means storing billions of 32-byte binary strings. One for each participant you intend to communicate with. They must also do this for every possible device they wish to communicate with, otherwise others can just eavesdrop on their conversations if keys are reused.
You've turned an O(n) space problem into an O(n^2) space problem.
If your n is small, you might still be tempted to make the move, but it will quickly become unmanageable as n grows.
Furthermore, if a devices loses their key, you must:
1. Generate/share a new keypair with n devices (which at Internet scale is billions), and
2. Do so in a way that doesn't open the door for backdoors, which probably requires digital signatures and therefore you're not actually getting rid of public-key cryptography
I think at Internet scale, we're still years away from having reliable post-quantum cryptography to even recommend let alone adopt, but that's going to be the way to go. Something something baby and bathwater.
Storing, distributing, and synchronizing billions of small blobs is actually a solved problem and not that big a deal. I wouldn't be so quick to write it off, because it has a real, working, easily understood invalidation and revocation story, which PKI does not.
A system that combines asymmetric crypto public-private key pairs with the distributed-whitelist properties of simple passwords, by treating the public keys as PSKs, has a lot of compelling benefits.
Not for HTTPS, which is what we're talking about.
For WPA2-PSK, okay, you have a small n, it might work.
For user authentication against a local database, you're not encrypting, you're using validating against a password hash and then elevating the user's privilege for the duration of their session. That's far removed from what we mean when we say PSK (pre-shared keys), where encryption is implied.
However, distributing a PSK to every device is not friendly. 10 Million Passwords will at minimum take about 80 MB, some phones and IoT devices have only ten times of that as permanent storage or less.
Passwords work mostly because the other side has the storage and cpu capacity to operate with them
I think the bigger issue with PSK at scale is key generation and distribution.
Although I'm unfamiliar w/ anything more sophisticated than a password manager like 1Password and S/Key - I'd like to see if I could have a go with this for my personal kit.
Can anyone recommend keywords/papers that talk about pre-shared key authentication at automated scale?