Idea: A Generic P2P Network Client

Every now and then one has a half-baked idea about some project that they aren’t likely to be able to do because of lack of time. I’ve written about such random app ideas before, but they were mostly about small apps.

Here I’d like to share an idea for something a bit bigger (and therefore harder to spare time for) – a generic P2P network client. P2P networks are popular in various domains, most notably file sharing and cryptocurrencies. However, in theory they can be applied to many more problems, social networks, search engines, ride sharing, distributed AI, etc. All of these examples have been implemented in p2p context, and they even work okay, but they lack popularity.

The popularity is actually the biggest issue with these applications – in order to get a service to be popular, in many cases you need a network effect – a p2p file sharing with 100 users doesn’t benefit from being p2p. A social network with 100 users is, well, useless. And it is hard to get traction with these p2p services because they require an additional step – installing software. You can’t just open a webpage and register, you have to install some custom software that will be used to join the p2p network.

P2P networks are distributed, i.e. there is no central node that has control over what happens. That control is held over the binary that gets installed – and it’s usually open source. And you need that binary in order to establish an overlay network. These networks reuse the internet’s transport layer, but do not rely on the world wide web standards, and most importantly, don’t rely heavily on DNS (except, they actually do when run for the first time in order to find a few known seed nodes). So once you are connected to the network, you don’t need to make HTTP or DNS queries, everything stays in the specifics of the particular protocol (e.g. bittorrent).

But the fact that not only you have to trust and install some piece of software, you have to be part of the network and exchange data regularly with peers. So it really doesn’t scale if you want to be part of dozens of p2p networks – each of them might be hungry for resources, you have to keep dozens of applications running all the time (and launching on startup).

So here’s the idea – why don’t we have a generic p2p client. A software that establishes the link to other peers and is agnostic on what data is going to be transferred. From what I’ve seen, the p2p layer is pretty similar in different products – you try to find peers in your immediate network, if none are found, you connect to a known seed node (first by DNS which uses DNS round-robin and then by a list of hardcoded IPs), and when you connect the seed node gives you a list of peers to connect to. Each of those peers has knowledge of other peers, so you can quickly be connected to a significant number of peer nodes (I’m obviously simplifying the flow, but that’s roughly how it works).

Once you have an established list of peers, you start doing the application-specific stuff – sharing files, downloading a cryptocurrency ledger, sharing search indexes, sharing a social network profile database, etc. But the p2p network part can be, I think, generalized.

So we can have that generic client and make it pluggable – every application developer can write their own application ontop of it. The client will not only be a single point for the user, but can also manage resources automatically – inbound and outbound traffic, CPU/GPU usage (e.g. in case of cryptocurrency mining). And all of these applications (i.e. plugins) can either be installed by downloading them from the vendor’s website (which is somewhat similar to the original problem), or can be downloaded from a marketplace available within the client itself. That would obviously mean a centralized marketplace, unless the marketplace itself is a p2p application that’s built-in the client.

So, effectively, you’d be able to plug-in your next file sharing solution, you next cryptocurrency, encrypted messaging, or your next distributed social network. And your users won’t have the barrier of installing yet another desktop app. That alone does not solve the network effect, as you still need enough users to add your plugin to their client (and for many to have the client to begin with), but it certainly makes it easier for them. Imagine if we didn’t have Android and Apple app stores and we had to find relevant apps by other means.

This generic client can possibly be even a browser plugin, so that it’s always on when you are online. It doesn’t have to be, but it might ease adoption. And while it will be complicated to write it as a plugin, it’s certainly possible – there are already p2p solutions working as browser plugins.

Actually, many people think of blockchain smart contracts as a way to do exactly that – have distributed applications. But they have an unnecessary limitation – they work on data that’s shared on a blockchain. And in some cases you don’t need that. You don’t need consensus in the cryptocurrency sense. For example in file sharing, all you need to do is compute the hash of the file (and of its chunks) and start sending them to interested peers. No need to store the file on the blockchain. Same with instant messaging – you don’t need to store the message on a shared immutable database, you only need to send it to the recipients. So smart contracts are not as generic solution as what I’m proposing.

Whether a generic client can accommodate an unlimited amount of different protocols and use cases, how would the communication protocol look like, what programming languages it should support and what security implications that has for the client (e.g. what’s the sandbox that the client provides), what UI markup will be used, are all important operational details, but are besides the point of this post.

You may wonder whether there isn’t anything similar done already. Well, I couldn’t find one. But there’s a lot done that can support such a project: Telehash (a mesh protocol), JXTA (a p2p protocol) and its Chaupal implementation, libp2p and Chimera (p2p libraries), Kademlia (a distributed hash table).

Whether such a project is feasible – certainly. Whether its adoption is inevitable – not so certainly, as it would require immediate usefulness in order to be installed in the first place. So, as every “platform” it will face a chicken-and-egg problem – will people install it if there are no useful plugins, and will people write plugins if there are no user installations. That is solvable in a number of ways (e.g. paying developers initially to write plugins, bundling some standard applications (e.g. file sharing and instant messaging). It could be a business opportunity (monetized through the marketplace or subscriptions) as well as a community project.

I’m just sharing the idea, hoping that someone with more time and more knowledge of distributed networks might pick it up and make the best of it. If not, well, it’s always nice to think about what can the future of the internet look like. Centralization is inevitable, so I don’t see p2p getting rid of centralized services anytime soon (or ever), but some things arguably work better and safer when truly decentralized.

The post Idea: A Generic P2P Network Client appeared first on Bozho's tech blog.