Peer-to-peer architecture research

I am working on a private sharing app. I need recommendations for a peer-to-peer architecture. Ideally, I don’t have to build the peer-to-peer infrastructure. I really just want to build a great app on top of one.

Use case

• A person’s data is stored primarily on their phone.
  • A person’s data can be backed up on their computer or perhaps cloud storage of their choice.
• A person will not have more than 250 friend peer connections.
  • It’s ok for people to invite each other to connect using an out-of-bands method (email, text message, in person, et al). An ugly identifier, like a public key string, is ok.
• A person’s data is shared selectively with another person (a friend). For example, the person might have some posts that are meant for all friends, some friends, or just themself.
  • There is no public sharing. Data does not have a public URL because it’s only ever sent from its creator to the creator’s friends.
• A person’s data can be selectively requested by their friends. For example, a friend should not have to store everything a person has sent to them forever. The friend can cache the data, purge the cache, and re-request the data in the future from the person.
• A person can delete their data and request their friends’ devices also delete the data.
• Ideally, a person’s data can only be obtained from the person (or a system the person has authorized to act on their behalf), not from friends the person has sent their data to. No relaying a person’s data by a mutual friend or friend-of-a-friend.
  • This is not a hard requirement. Perhaps peers could help with the offline problem?
  • It’s ok for data to not be available immediately, but it’s not ideal. If a person’s friend is offline, perhaps requests are queued until the person and friend are both online. Perhaps both a person’s phone and their backup computer could respond to requests from friends.
• The role of a server not operated by the person should be minimal.
  • It could provide an assist for connecting people.
  • It could provide temporary caching or relaying of data if done in a way the server operator could not inspect the contents of the data
  • Ideally whatever the server must do is able to be done by other independently run servers as well.
• The app is a progressive web app. The web app only contains the business logic. User data is never stored by the server hosting the web app. Any webserver could host the web app and people could connect their locally stored data to it. Preferably, the web app does not need to be wrapped in a native OS app wrapper to function.

Architecture thoughts

• No immutable data structures. When a person deletes their data, it should be as if the person never stored the data on their device.
  • A person’s control of their data is more important than censorship resilience. Some protocols optimize for replication durability. I believe people are more likely to regret sharing something more than they are to be subjects of state-sponsored censorship.
    • This is one reason why I don’t think IPFS is helpful for this use case. If you think it could be, please educate me.
• I don’t think blockchains or cryptocurrencies are helpful to this use case. If you think they could be, please educate me.
• I have a strong bias against GraphQL, but will tolerate it if necessary.
• I wish Solid were better and usable by normal people. I can imagine people using my app with their data being stored in their pod and their friends having per-user access permissions to access certain files. If a person had an always-online device within their control like this, that would simplify things.
  • I can imagine something like a server that runs as a proxy on top of people’s Google Drive, Dropbox, Fastmail files, Proton cloud storage as an interim solution. It would require high trust of the proxy server provider.
  • Asking people to sign up for a new cloud storage provider is a high barrier to entry and not one that I think is acceptable for my app.
• I can imagine something like Signal’s server architecture working as an offline intermediary. When both parties are online, they communicate directly. When one party is offline, requests for the other party’s data are sent to an intermediary. The Signal protocol allows for anonymity of the sender, receiver, and message contents. When the offline party comes back online, they check for new requests and respond with the data. It’d be like HTTP request and responses being facilitated by an intermediary for when either the requester or responder are offline.
  • I think the Matrix protocol could also be implemented for use in this way. The one aspect I am uncertain about is clearing the messages from the server once retrieved. Signal’s server only ever temporarily stores messages.
  • A goal is to avoid a centralized, controlling server. Multiple servers could be operated for this purpose. I assume this would mean the app would have to check a list of servers for requests/responses and the user could manage which ones they wanted to use.
• I can imagine each person being their own Tor Onion Service. This would only allow data exchange when both parties are online. I am not sure if an Onion Service can be run from the browser, but I do know running the Tor Snowflake proxy works in the browser.

Things I’ve heard about but not fully investigated

If you have any knowledge or experience with these, please let me know how they may or may not be useful for the use case described above.

• Any-sync
• ODD
• p2panda
• Socket Runtime
  • Unclear if the P2P component can be used on just the Web
• Spritely

Have ideas?

Please send me your thoughts, knowledge, things you’ve heard about, or anything else you think might be helpful.

You can respond to this post on the social web or email me.

I will update this post as I learn!