I'm back from devcon3 and we discussed all the things! I spoke about my vision for devp2p in 2018 (slides here). I mentioned libp2p integration options in the talk and also argued for moving toward libp2p in follow-up discussions. Naturally there were quite a few questions about that. This post is supposed to answer those.
libp2p is a core component of the ipfs project. It's a collection of libraries that are supposed to help with implementing peer-to-peer systems. People often ask: why doesn't Ethereum use libp2p?
The answer is two-fold:
- devp2p and libp2p aren't directly comparable because devp2p is a system and libp2p is a framework. You can assemble a system out of libp2p's components, but the result might not necessarily be compatible with another system that uses different components.
- libp2p didn't exist when we started building the peer-to-peer networking stack for Ethereum. There were discussions about building something together very early on, but in the end we were more set on shipping a working system than to discussing how to make the most flexible and generic framework.
Now that devp2p has been in production use for two years, I feel it's time to revisit this topic. libp2p has come a long way too, with two interoperable implementations, a couple of transports and production use in ipfs.
I've been thinking about ways to integrate libp2p for quite a while, but couldn't figure out how to make use of it in a way that fits. My initial goal was to join their DHT, which turned out to be impractical. There's a much easier thing we can do though: use libp2p's transports and multiplexing algorithm. This saves us a bunch of code and gives us a few new protocols (like uTP) that we can try out without too much work.
Using the transports makes something else a lot easier: there's a project that tries to make Ethereum blockchain data available and linkable through ipfs (via ipld). Similarly, Swarm (the incentivized Ethereum file system) would be suitable as a backend store for ipfs data because it solves the availability problem for files that few people are interested in. It'd be awesome if ipfs nodes could just speak to the Ethereum network to get blockchain data and Swarm files.
Short Term Plan
The work to integrate libp2p starts now:
There will be a version of Whisper v6 that runs on libp2p. We've chosen Whisper as our experimentation vehicle because it's been lumping along without much use. Whisper peering sucks on devp2p because nobody runs it.
There will also be a separate experiment to run the existing 'eth' application protocol (the one that relays blocks and transactions) on libp2p. We're not expecting this to become canonical anytime soon. It'll still be interesting to see whether we can get additional performance out of the multiplexing, though.
Long Term Plan
The ultimate goal is to adopt libp2p's transport protocols as the way Ethereum nodes communicate. For this to work, those protocols need to be available to all relevant Ethereum node implementations. This will take time.
We have a transition plan: future versions of the devp2p Node Discovery protocol will include 'node records' in which multiple transports can be listed. The proposed format for these records is ENR (Ethereum Node Records).
Once ENR is adopted and devp2p's application layer works on top of libp2p transports, we can start using the transports between compatible implementations. Everyone else can continue to use RLPx until we decide to turn it off.