On Wed, 17 Jun 2020 at 15:52, degski
I have a *PoC* of a sax::aes_random_device at https://github.com/degski/aes_random_device .
It is a PoC and also needs research, it is undoubtedly much better than mt19937, and on Windows/MSVC is faster (~30%) than std::mt19937. The Crypto-claim needs more code and research related to backtracking-resistance ...
The back-tracking resistance is only relevant in a cryptographic context where the [this] CPRNG is under attack, with if I understood it well, unlimited access to streams [blocks] of data and unlimited time [as an assumption, i.e. somebody using a data-centre to actively crack a CPRNG]. The CPRNG is fully seeded with std::random_device, whereafter the device needs no more entropy, it is 'spreading out' that initial entropy. The entropy is massive as compared to a PRNG (mt19937 a.o.). degski