I am making HTTP calls on the client side and the server expects me to supply a client nonce. It needs to be 4 bytes long. I am planning to use this combination:
base64(parts of the MAC address + random bits generated by RNG)
Given that the protocol is created by experts (thus its security is not to be questioned), is my nonce calculation acceptable in your view ?
From a cryptographic protocol perspective what is usually important about a nonce (number used once) is its one-timeness. Having it generated with a (P)RNG is rather a matter of convenience. If a cryptographic protocol has specific entropy requirements on a nonce or assumes it to be otherwise cryptographically secure in some way - that would be typically known, communicated or stated upfront, because that isn't a standard expectation.
I am making HTTP calls on the client side and the server expects me to supply a client nonce
This is a typical scenario for the prevention of a replay attack - relying on the one-timeness of certain parts of the client input.
I am planning to use this combination: base64(parts of the MAC address + random bits generated by RNG)
When you rely on parts of the MAC address as an input to a linear transformation, you are setting yourself for an increased probability of the nonce repeating, increasing the chance of a successful replay attack, or alternatively you increase the chance of an adversary guessing your next nonce in advance, making her job easier in a sense.
Notwithstanding, in an expert chosen cryptographic protocol (as assumed in the question) it will be more common for both parties to choose a nonce, mitigating inadequate choice of a nonce by only one party. With a server nonce and a client nonce, establishing a communication frequently looks like this,
and when you don't chose cnonce adequately, or don't have a client nonce at all, an adversary can observe f(snonce, challenge) responses, where all the inputs are chosen by it, when it masquerades as a server. This can at times deteriorate to a chosen plaintext attack.
I suggest that you simplify the nonce to be along the lines of,
cnonce = base64(random bits generated by RNG)
where you RNG has sufficient entropy. And if you need repeatability for test purposes, eg if you run your protocol client side tests in a farm on different boxes you can:
seed(RNG, parts of the MAC address)
cnonce = base64(random bits generated by RNG)
That way your tests on every box will be amenable for debugging.
