On Tue, Dec 06, 2022 at 12:39:40AM -0500, Peter Todd via bitcoin-dev wrote:
> 10 or 20 nodes is completely meaningless. Pools run nodes themselves, which by
> default connect to 8 outgoing peers. There's about 5000 IPv4 listening nodes on
> the network. When a node learns of a new block, it tells all it's peers that
> the new block exists.
> 
> For your censorship to work, there has to be a substantial propability that a
> miner *only* runs a single node (they don't), that has no incoming peers, and
> all 8 peers of that node happen to be one of your 20 censoring nodes.
> Obviously, since the probability of a given peer being a censoring node is
> 20/5000, all 8 being censored is extraordinarily unlikely.
> 
> Even if you ran so many nodes that 20% of the entire network was censoring, the
> probability of all 8 outgoing peers being censors is only 0.2^8 = 0.000256%
> 
> 
> This is an example of information being hard to censor and easy to spread. In
> fact, for full-rbf this same math works in our favor: for a node to have a 50%
> chance of connecting to at least one full-rbf peer, just 8.3% of the network
> needs to run full-rbf. 5000 IPv4 nodes * 8% = 400 nodes.
> 
> The percolation threshold doesn't need to be met for this to be succesful,
> because someone to just run a full-rbf node that connects to every single
> listening node simultaneously.

FYI here's a percolation simulator for full-rbf:

https://github.com/mzumsande/fullrbf_simulation

It finds similar results to my math above.

-- 
https://petertodd.org 'peter'[:-1]@petertodd.org