Are the connections between the nodes in Marlin network visible to all globally or just known locally?
We are currently experimenting with models that involve both open and closed topologies. The tradeoffs as far as I understand are:
- Easier to deanonymize
- Easier to inflict eclipse attacks
- If on-chain identities and IP addresses are linked, easier to bribe or grief victims as revenge
- Can not be sure if the network is really decentralized
- Security by obscurity
- Super nodes might exist undetected that can easily censor or partition the network
- Determined actors can figure connections anyway (for example, https://orbilu.uni.lu/bitstream/10993/42573/1/43.pdf, https://link.springer.com/chapter/10.1007%2F978-3-030-32101-7_32 – I think their presentation in Scaling Bitcoin goes over through these tradeoffs in detail)
Tools like CoinSpace already enable network scanning . I believe even TxProbe  uses the same. Preventing emergence of influential/super nodes in open networks is impossible. If a decentralized system assumes the existence of adversarial actors trying to get control over the network and/or makes running a node even marginally profitable, elements of centralization is more or less a given. Better to have them public rather than build a Maginot Line that can be easily breached by a determined actor.
In fact, Miller’s paper concludes with-
“We hope that the findings in this paper—that understanding topology can identify structural faults
to the broadcast—will encourage the Bitcoin community to evolve the protocol to explicitly support efficient topology discovery.”
I agree with whatever’s been shared above. I also like the question and think that there should be wider discussion on the topic. I suspect that a choice between structured and unstructured p2p networks would be of relevance too. Its astonishing how little is spoken about p2p networking while we work in the domain of p2p networks.
Another relevant work: http://dsn.tm.kit.edu/publications/files/323/bitcoin_timing_analysis_dsn.pdf