My understanding is you only need 33% of the hash power at any given time. Since PoW is only done as part of sending transactions, it probably takes less hash power than you'd think to cause problems.
>Why does everyone repeat that Byzantine consensus requires maximum 33% of participants to be dishonest?
Not 33% of participants, 33% of the hash power, could just be one participant with a pile of GPUs or ASICS or "JINN" chips lol. That's the claim made by the IOTA author, anyway.
Right now it wouldn't surprise me if someone could amass 90+% of IOTA hash power anyway.
Okay but it was more of a general question. I see Hashgraph and others always saying that they need 33% of participants to be honest. But with unforgeable message signatures that limitation doesn't apply.
Virtual Voting in hashgraph requires a 2/3 agreement.
Of course PoW provides some protection against sybil attacks, but the reality is that with enough hashpower the network can be overtaken. (Hence why HashGraph is a closed network.)
First, a paper written three weeks ago of an unproven currency is a bit of stretch.
Second, I didn't say it was necessary to order transactions, I said that is what it is used for, which is correct. You are replying to a point that I didn't make.
If messages cannot be forged, then a consensus could make positive progress with even 99% of participants being dishonest.