@vitalik.eth

Increasing bandwidth is safer than reducing latency With PeerDAS and ZKPs, we know how to scale, and potentially we can scale thousands of times compared to the status quo. The numbers become far more favorable than before (eg. see analysis here, pre and post-sharding https://vitalik.eth.limo/general/2021/05/23/scaling.html ). There is no law of physics that prevents combining extreme scale with decentralization. Reducing latency is not like this. We are fundamentally constrained by speed of light, and on top of that we are also constrained by: * Need to support nodes (especially attesters) in rural environments, worldwide, and in home or commercial environments outside of data centers. * Need to support censorship-resistance and anonymity for nodes (especially proposers and attesters). * The fact that running a node in a non-super-concentrated location must be not only possible, but also economically viable. If staking outside NYC drops your revenues by 10%, over time more and more people will stake in NYC. Ethereum itself must pass the walkaway test, and so we cannot build a blockchain that depends on constant social re-juggling to ensure decentralization. Economics cannot handle the entire load, but it must handle most. Now, we can decrease latency quite a bit from the present-day situation without making tradeoffs. In particular: * P2P improvements (esp erasure coding) can decrease message propagation times without requiring individual nodes to have lower bandwidth * An available chain with a smaller node count per slot (eg. 512 instead of 30,000) can remove the need for an aggregation step, allowing the entire hot path to happen in one subnet This plausibly buys us 3-6x. Hence, I think moderate latency decreases, to a 2-4s level, are very much in the realm of possibility. But Ethereum is NOT the world video game server, it is the world heartbeat. If you need to build applications that are faster than the heartbeat, they will need to have offchain components. This is a big part of why L2s will continue to have a role even in a greatly scaled Ethereum (there are other reasons too, around VM customization, and around applications that need _even more scale_). Ultimately, AI will necessitate applications that go faster than the heartbeat no matter what we do. If an AI can think 1000x faster than humans, then to the AI, the "subjective speed of light" is only 300 km/s. Hence, it can talk near-instantly within the scope of a city, but not further. As a result, there will inevitably be AI-focused applications that will need "city chains", potentially even chains localized to a single building. These will have to be L2s. And on the flipside, it would be too much of a cost to make it viable to run a staking node on Mars. Even Bitcoin does not strive for this. Ultimately, Ethereum belongs to Terra, and its L2s will serve both hyper-localized needs in its cities, and hyper-scaled needs planet-wide, and users on other worlds. Milady.

16 replies

25 recasts

167 reactions