> Only DPOS BFT can efficiently scale to an unlimited number of validators (at cost to latency). Other protocols grow the resource requirements for finality with O(N²) the number of participants as everyone must talk to everyone twice for each block or checkpoint.
There’s a large class of protocols, including traditional BFT algos, where the overhead to each node is proportional to the number of participants. In this proposed DPOS BFT too, it’s the case that “everyone must talk to everyone twice”: with N validators, you get finality after 2N blocks, and each of these blocks needs to be broadcasted to all N participants, so that’s 2N² bilateral communications total.
> The end result is 10x faster than Casper for finality for individual blocks, but a new block reaches finality every two seconds versus every 30 minutes.
The proposed algo achieves this because of lower node count (21) and higher overhead, not because of any fundamental algorithmic improvement.
I strongly recommend this article I wrote earlier this year that explains the full tradeoff space: https://ethresear.ch/t/in-favor-of-forkfulness/1225
Author: Vitalik Buterin