You know, I didn’t have that in mind at all, but my subconscious must have guided me to the analogy. It just fits too well.
Mr. President, we mustn’t have a seashell gap!
Mein Fuerher! I can walk!
Neat. It appears the algorithm depends on non-colluding parties, though there are later extensions that improve that.
The name Leslie Lamport sounded familiar. I realized I’d read this article recently:
Collusion is always a potential problem with distributed systems—as an extreme example, if there are only 3 parties, and you know you have $50 in your account, but the other two are dishonest and reply to any queries that you have $0, what is the system (you, as the only honest user) as a whole supposed to do? They can boot you out as a malfunctioning node, rather than the other way around. There may be more detailed records, and cryptography can make it unreasonable for users to impersonate others or forge their signatures, but this is hardly a trivial problem.
Anyway you have to worry about general Byzantine failures, not addressed in Lamport’s original description but later considered by others. As well as protocol efficiency and resistance to various forms of attack, if there is to be an actual deployment.
On Wikipedia some production uses of Paxos (and other consensus) algorithms are named, but they do not name any specific financial or payments application, just general storage and database products.
Without thinking about it too hard, my suspicion is that you can either optimize for collusion or not. With collusion, the best you can do is resist a 49% attack. Without collusion, then you can get away with arbitrarily low agreement; perhaps only 2 nodes out of N, if the probability of an accidental collision is low enough. If some other assumptions are made (like that the successful receipt of a message is enough to assume correctness), then maybe just one node is enough.
I just watched the NOVA episode a couple nights ago. I couldn’t quite decipher it either, although they certainly tried. I DO finally know what “blockchain” is, because it was finally defined in layman’s terms.
In short, if you don’t understand it, you probably shouldn’t invest in it.