10% of Ethereum nodes at risk of being booted from Hetzner(web3isgoinggreat.com)
web3isgoinggreat.com
10% of Ethereum nodes at risk of being booted from Hetzner
https://web3isgoinggreat.com/?id=10-of-ethereum-nodes-at-risk-of-being-booted-from-cloud-hosting-provider
8 comments
Bitcoin doesn't make outbound connections to multiple hosts sharing a /16 (there is some analogous prefix for IPv6 but I don't recall what it is).
If you provide it with a file containing a prefix trie that maps IPs to ASNs (or "enclosing ASN"-- e.g. merging any non-multihomed AS into its parent) then it won't make multiple outbound connections to the same ASN. There was some aspirations at some point of shipping a map file with the software, but I think the people working on that stopped working on Bitcoin.
Gold standard would be if you were multihomed yourself, constructing such a map from your own routing tables that collapsed every ASN into its distinct parent such that no outbound peers share a common network at all. (Bitcoin nodes don't make many outbound connections, so this wouldn't be totally crazy). Though that would still be a little approximate because there is no guarantee that interdomain routing is symmetrical.
If you provide it with a file containing a prefix trie that maps IPs to ASNs (or "enclosing ASN"-- e.g. merging any non-multihomed AS into its parent) then it won't make multiple outbound connections to the same ASN. There was some aspirations at some point of shipping a map file with the software, but I think the people working on that stopped working on Bitcoin.
Gold standard would be if you were multihomed yourself, constructing such a map from your own routing tables that collapsed every ASN into its distinct parent such that no outbound peers share a common network at all. (Bitcoin nodes don't make many outbound connections, so this wouldn't be totally crazy). Though that would still be a little approximate because there is no guarantee that interdomain routing is symmetrical.
Love this; Another factor could be peer latency.
You know anybody looking to flesh it out into an idea?
> "16% of all hosting nodes (a category that makes up 62% of all nodes by network type) are hosted with Hetzner—10% of all nodes. If 10% of all Ethereum nodes being supported by one company sounds awfully centralized to you, wait til you hear that 30% run on Amazon services."
A real problem for decentralization proponents: there's always a positive return to scale somewhere, which will tend to suck everything back into a center. So you get "decentralization" which is just labels over AWS. Or Bitcoin converging on a shrinking number of pools.
A real problem for decentralization proponents: there's always a positive return to scale somewhere, which will tend to suck everything back into a center. So you get "decentralization" which is just labels over AWS. Or Bitcoin converging on a shrinking number of pools.
Decentralization of resources is different from decentralization of ownership. What web3 is pushing for is decentralization of ownership. So even if people are using AWS, they still have control over what they run.
Now Hetzner might be trying to undermine this by identifying and shutting down Eth nodes on their servers. But I imagine the next step for Web3 would be to make it harder to identify these nodes.
Now Hetzner might be trying to undermine this by identifying and shutting down Eth nodes on their servers. But I imagine the next step for Web3 would be to make it harder to identify these nodes.
This is a thought experiment:
hypothetical scenario, pretend our fictitious p2p network is in a vacuum and all peers are speaking over ipv4.
How would it affect peer diversity if each peer refused to add an other remote peer that shares any of its IP octets? Obviously ip6 would break this to an extent. What if there were other stipulations, for example where nodes will not peer-exchange with a peer that is under X latency, or sharing any octets, or under N number of hops, rejecting signs of being geo-proximal?
Kademlia, Tapestry and other p2p schemes don't really address this as much as one might hope to. I know the practice of maintaining node diversity/non-diversity is there in the industry, but it seems as a practice to have faded out when the last of the good malware and p2p devs retired.