Re: Weakness of Feistel ciphers

In article <7xodyx74on.fsf@xxxxxxxxxxxxxxxxxxx>,
Paul Rubin <> wrote:
kim@xxxxxxxxxxx (Kim G. S. Øyhus) writes:
It has to work modula a Mersenne prime, product of primes, or similar.

OK. It sounds like you're asking for a bijection from Z//p to Z//p
where you can quickly compute both the encryption and its inverse. Is
that right?

Yes, and perhaps other fields or semigroups as well, as you guessed.

That is straightforward to do using AES or SHA or whatever as a
building block. Schroeppel's paper about the Hasty Pudding Cipher
explains how.

Thank you for your advice. I will examine that.

I am not a rookie. I have worked professionally with crypto for over
4 years now, and have made stuff like the fastest RSA for the ARM
processor, and a system which converts fingerprints to crypto keys.

The kinds of questions you're asking make it sound like you may
understand how to implement math algorithms, but you don't actually
understand crypto. There's nothing wrong with that, but it means
you're not yet in a position to roll your own.

Since I do not need strong crypto, I think it is perfectly all right
to make my own. I just need something that is strong enough, stronger
than ordinary pseudo random generators. My tests say I have achieved
that, but they can be wrong.