Skip Navigation

Distributed Hosting

Hi all - So I posted about distributed hosting yesterday. I wrote up some thoughts on how I think it could work, and I'm planning to start work on it -- if anyone has feedback on my proposal, or wants to get involved to help, 100% let me know as I'd love to hear.

(Edit: removed the link from the URL field, as it pasted it unformatted into the post which is not productive. Click the link in the paragraph above if you want to read in an un-eye-crossing format.)

9

You're viewing a single thread.

9 comments
  • Please add some text formating, at least some line breaks ...

    • I asked ChatGPT to do that for you :)


      So I made this forum to work on one specific piece of software that I think could benefit Lemmy (and the overall fediverse community) substantially. I’ll lay out what I want to make and why, in some detail.

      I apologize for the length, but I can’t really do this without some level of support and agreement from the community, so hopefully the wall of text is worth it if it resonates with some people and they’re swayed to support the idea. If something like this already exists please let me know. I looked and couldn’t find it, which is why I’m making this extensive pitch about it being a good idea. But, if it’s already in the works, I’d be just as happy working on existing tech instead of reinventing it.

      The Problem

      In short, the problem is that you have to pay for hosting. Reddit started as a great community, just like Lemmy is now, but because it was great it got huge, which meant they had to pay millions of dollars to run their infrastructure, and now all of a sudden they’re not a community site anymore. They’re a business, whether they like that or not.

      Fast forward fifteen years and look how that turned out. I think this will impact Lemmy in the future, in very different ways but still substantially. It’s actually already, at this very early stage, impacting Lemmy: There are popular instances that are struggling under the load, and people are asking for donations because they have hosting bills.

      Sure, donations are great, and I’m sure these particular load problems will get solved – but the underlying conflict, that someone who wants to run a substantial part of the network has to make a substantial financial investment, will remain. Because of its federated nature, Lemmy is actually a lot better positioned to resist this problem. But, it’ll still be a problem on some level (esp. for big instances), and wouldn’t it be better if we just didn’t have to worry about it?

      The Solution

      Basically, I propose that all users help run the network. Lemmy is a big step forward because a lot more of users can help than before, but even in Lemmy, only a small fraction of people will choose to make instances, and you’ll still have big instances serving lots of content.

      I propose to make it trivially easy for the end-users to carry the load. They can install an app on their phones, or a browser plugin, or run something on their home computer, but they have absolutely trivial ways to use their hardware to add load capacity. The load on the instances will be way reduced just from that option existing, I think.

      I would actually argue for taking it a step further and having instance operators be able to require load-carrying by their users, but that’s a choice for the individual operators and the community, based on observation of how this all plays out in practice.

      One Implementation

      It’s easy to talk in generalities. I’m going to describe one particular way I could envision this being implemented. This proposed approach is actually not specific to Lemmy – it would benefit Lemmy quite a lot I think, but you could just as easily use this technology to carry load for a Mastadon instance or a traditional siloed web site. It’s complementary to Lemmy, but not specific to it.

      Also, this is going to be somewhat technical, so feel free to just skip to the next section if you’re just interested in the broad picture.

      So like I said, I propose to make peer software that provides capacity to the system to balance out the load you’re causing as an end-user. The peer is extremely simple – mostly it runs a node in a shared data store like IPFS or Holepunch, and it serves content-addressable chunks of data to other users.

      You can run it as an app on your phone if you have unlimited data, you can run it as a browser plugin (which speeds up your experience as a user, since it’ll have precached some of the data the app will need), you can run it on your computer back at home while you access Lemmy from the road, etc.

      The peer doesn’t need to be trusted (since it’s serving content-addressable data that gets double-checked), and it doesn’t need to be reliable or always on. The system keeps rough track of how much capacity your peer(s) have added, and as long as it’s less then your user has consumed, you’re fine if your peer goes away for a couple of days or something.

      When you, as a user, open your Lemmy page served by the instance, what you get served back is tiny: Just a static chunk of bootstrapping javascript, a list of good peers you can talk to, and a content hash of the “root” of the data store

    • Part 2:

      (Continued from the post)

      What's the Next Step?

      I started touching on some imagined future steps, but this chunk is already a plenty big and ambitious thing. So, here's an initial plan for how I want to attack taking first steps and bring myself into contact with the engineering reality (as opposed to the rosy broad picture). Hopefully at the end of this chunk of work, the vision will have adapted somewhat to the reality of what's useful, what's possible, what the community's feedback is, what the issues and problems involved are, etc.

      (And, obviously, I want to communicate with the Lemmy devs to make sure these ideas are in line with their vision. I'm laying this all out so extensively partly so that the community has a full explanation of what I'm proposing to do and why.)

      So, first steps: I'm making a Lemmy instance that I can use for implementing this. I'm waiting for my hosting to go up so I can make it live, but once it's up, I'll start working on it + posting from the testbed about what's going on. My initial coding task list is:

      • Set up the peer software with the content-addressable store

      • Start to have my instance do peer discovery, make the app that runs in people's browsers from my instance become more AJAX-y and begin to request data from the peers instead of the instance.

      • Once that part's working on my instance, I'd aim to be able to move pieces of the actual app onto the peers -- construct the bootstrap code, continue the AJAX-ification of the code on my Lemmy instance, and have the bootstrapping app construct the end-user application directly from data from the peers.

      • Start to tackle the browser app making updates to the data store via requests to the peers, which will involve a lot of work and lot of sorting out replication issues, security and trust issues, and performance issues.

      That's already a fairly large amount to take on. I have further ideas about how the system could move forward from there, but even just that represents (1) an ambitious thing to tackle (2) significant proposed changes to the instance software (3) if it works, a fantastically useful tool that instance operators could use to reduce their instance load if they want to. So, I'm limiting the plan to that much for now until I get some contact with the technical reality and with the community.

      What You Can Do

      So if you've read to the end, maybe you think this is a good idea. Want to help? This is a bunch of work already and I'd love it if people wanted to help get it done. Leave a comment, let me know what you think whether positive or negative, and if you want to help, 100% reach out and let's get it done. I'm skilled with software engineering in general, but I'm actually not too familiar in particular with web backends and AJAX, so someone more skilled than I am could probably help this along in a huge way. Specific things that might be useful:

      • If you want to run a peer or instance and help test the system

      • If you can help with coding

      • If you have feedback on these ideas in general, either positive or else things I've overlooked or need to adjust

      Hope to hear from you and thank you for reading my wall of text. Let me know what you think + cheers to you.

      • I got a spare raspberry pi set up as a server. I can use that to host stuff and am okay in programming (not rust though). Let me know if I can be of assistance in anyway. Be happy to help with this effort

        • Yes, absolutely! Sorry for the silence... I was working on code with a little bit expanded scope from the original project, but it's shaping up to be maybe within a week or two something that could actually be tested. You can read the update about the current state of things; in it I talk about having a test instance set up, and wanting to set up proxy caches for it to be able to test the whole system in real-world functioning. The code's not ready yet, but maybe if in a couple weeks you still want to help with testing, I can help you get a proxy node set up on your Pi and then that can form part of the initial proof-of-concept on that testbed server?

          • For sure and no worries about the delay in response. I am out on vacation but when I get back, I can lend help set up a proxy node. Maybe set it up as a container on docker

    • Or, what the hell; here's a copy-paste of the whole thing to be able to read it here too. Part 1:

      Overview

      So I made this forum to work on one specific piece of software that I think could benefit Lemmy (and the overall fediverse community) substantially. I'll lay out what I want to make and why, in some detail. I apologize for the length, but I can't really do this without some level of support and agreement from the community, so hopefully the wall of text is worth it if it resonates with some people and they're swayed to support the idea.

      If something like this already exists please let me know. I looked and couldn't find it, which is why I'm making this extensive pitch about it being a good idea. But, if it's already in the works, I'd be just as happy working on existing tech instead of reinventing it.

      So:

      The Problem

      In short, the problem is that you have to pay for hosting. Reddit started as a great community, just like Lemmy is now, but because it was great it got huge, which meant they had to pay millions of dollars to run their infrastructure, and now all of a sudden they're not a community site anymore. They're a business, whether they like that or not. Fast forward fifteen years and look how that turned out.

      I think this will impact Lemmy in the future, in very different ways but still substantially. It's actually already, at this very early stage, impacting Lemmy: There are popular instances that are struggling under the load, and people are asking for donations because they have hosting bills. Sure, donations are great, and I'm sure these particular load problems will get solved -- but the underlying conflict, that someone who wants to run a substantial part of the network has to make a substantial financial investment, will remain.

      Because of its federated nature, Lemmy is actually a lot better positioned to resist this problem. But, it'll still be a problem on some level (esp. for big instances), and wouldn't it be better if we just didn't have to worry about it?

      The Solution

      Basically, I propose that all users help run the network. Lemmy is a big step forward because a lot more of users can help than before, but even in Lemmy, only a small fraction of people will choose to make instances, and you'll still have big instances serving lots of content. I propose to make it trivially easy for the end-users to carry the load. They can install an app on their phones, or a browser plugin, or run something on their home computer, but they have absolutely trivial ways to use their hardware to add load capacity. The load on the instances will be way reduced just from that option existing, I think. I would actually argue for taking it a step further and having instance operators be able to require load-carrying by their users, but that's a choice for the individual operators and the community, based on observation of how this all plays out in practice.

      One Implementation

      It's easy to talk in generalities. I'm going to describe one particular way I could envision this being implemented. This proposed approach is actually not specific to Lemmy -- it would benefit Lemmy quite a lot I think, but you could just as easily use this technology to carry load for a Mastadon instance or a traditional siloed web site. It's complementary to Lemmy, but not specific to it. Also, this is going to be somewhat technical, so feel free to just skip to the next section if you're just interested in the broad picture.

      So like I said, I propose to make peer software that provides capacity to the system to balance out the load you're causing as an end-user. The peer is extremely simple -- mostly it runs a node in a shared data store like IPFS or Holepunch, and it serves content-addressable chunks of data to other users. You can run it as an app on your phone if you have unlimited data, you can run it as a browser plugin (which speeds up your experience as a user, since it'll have precached some of the data the app will need), you can run it on your computer back at home while you access Lemmy from the road, etc. The peer doesn't need to be trusted (since it's serving content-addressable data that gets double-checked), and it doesn't need to be reliable or always on. The system keeps rough track of how much capacity your peer(s) have added, and as long as it's less then your user has consumed, you're fine if your peer goes away for a couple of days or something.

      When you, as a user, open your Lemmy page served by the instance, what you get served back is tiny: Just a static chunk of bootstrapping javascript, a list of good peers you can talk to, and a content hash of the "root" of the data store. What the bootstrapping code does, is to start at the "root" of what it got told was the current state of the content, and walk down from there through the namespace, fetching everything it needs (both the data and the Lemmy app to render it and interact with it) by making content-addressable requests to peers. Since it all started with a verified content hash, it's all trustable.

      It's important that the bootstrapping code in the browser verifies everything that it gets from every peer. You can't trust anything you get from the peers, so you verify it all. Also, you don't trust the peers to be available -- the bootstrapping code keeps track of which ones are providing good performance, and doesn't talk to just a single one, so if one is overloaded or suddenly drops out, the user's experience isn't ruined. Also, you're able to configure a peer you're running to always keep full a mirror of some part of the data store that you're invested in. That's vital, because this system can't magically make all data always available without anyone thinking about it -- it just decouples (1) an instance you can always reach, which is probably on paid hosting, from (2) a peer which provides the heavy lifting of load capacity, but might drop out at any time, i.e. can run on unmetered consumer internet. You as a moderator still need to ensure that (1) and (2) are both present if you want to ensure that your content is going to exist on the system.

      The end result of this is that the end-user's interaction with the system only places load on the instance when it first fetches the bootstrapping packge. My hope would be that it can be small enough that you can run a fairly busy instance on a $20/month hosting package, instead of paying hundreds or thousands of dollars a month. Also, like I said, I think culturally it would be way better if running a peer was a requirement to access the instance. That's up to the individual instance operators, obviously, but to me people shouldn't just be entitled to use the system. They have to help support it if they're going to add load (since it's become trivial enough that that's reasonable to ask). Aside from ensuring load capacity, I actually think that would be a big step up culturally -- look at the moderation problems every online forum has right now because people are empowered to come onto shared systems and be dicks. I think having your use of the system contingent on fulfilling a social contract is going to empower the operators of the system a lot. If someone's being malicious, you don't have to play whack-a-mole with their IP addresses to try to revoke their entitlement to be there -- you just remove their status as a peer and their privilege to even use the system you've volunteered to make available in the first place.

      I've handwaved aside some important details to paint the broad picture. How do updates to the content happen? How do you index the data or make it relational so you make real apps on top of this? How do you prevent malicious changes to the data store? How is a peer that's port-restricted or behind NAT still able to help? These are obviously not minor issues, but they're also not new or extraordinary challenges. This is already long enough, so I'll make a separate post addressing more of the nitty-gritty details.

      What's the Result?

      So to zoom back out: One result, hopefully, is that the experience becomes faster from the end-user perspective. Hopefully. I believe that the increase in capacity will more than make up for the slowness introduced by distributing the data store, but that's just theory at this point. I would also argue that this will start to open up possibilities like video streaming that are hard to do if instances host all the content. But regardless of that, I think big popular instances not having to pay ever-increasing hosting costs is huge. It's necessary. It's not a trivial benefit. And, in addition to that and the cultural issues, I think this improves the overall architecture of the system in one more very significant way:

      Because the Lemmy app itself becomes static (AJAX-utilizing javascript which exists fully within the shared data store), it becomes trivial to make your own custom changes to the app even if you don't want to run an instance. You can clone the Lemmy app in the data store, make revisions, and then tell the system that you want to see your same data but rendered with the new version of the web app. Ultimately the entire system becomes a lot more transparent and flexible from a tech-savvy user's perspective. You don't have to interact with "the Lemmy API" in the same way people had to interact with "the Reddit API" -- your modified or independent app just interacts directly with the data. This is a huge shift further in the same direction that started with federating the servers in the first place. Part of the further future beyond this document is the possibility of opening up a lot of tinkering possbilities for tech-savvy end users, and expanding what even non-techy end users would be able to do with the apps they're interacting with.

      Getting It Done

      So I think I'm hitting a length limit, so I'll fill in the details of the first steps I want to take, in the next comment.

    • So I actually didn't do that, Lemmy did :-). I posted in a different community, then posted the link to that post in /c/selfhosted, not realizing that me doing it that way would include an unformatted and awful-looking partial version of my post into this post. I've fixed it now. Just click the link in the text at the top to read the well-formatted version.

9 comments