Conversation

Notices

1. Embed this notice
   Eli Pariser (elipariser@mastodon.social)'s status on Saturday, 07-Jan-2023 00:34:56 JST Eli Pariser

   Ok continuing up the fediverse learning curve… what’s the best read on the debate around scalability questions for ActivityPub and federated systems?

   • Embed this notice
     Evan Prodromou (evan@prodromou.pub)'s status on Saturday, 07-Jan-2023 00:34:49 JST Evan Prodromou

     in reply to

     • Thomas Vander Wal

     @vanderwal @elipariser I'm interested in scalability. Most of the analyses I've read start with uninformed assumptions about the fediverse as a strawman and then draw unfounded conclusions from those bad bases.

   • Embed this notice
     Thomas Vander Wal (vanderwal@mastodon.social)'s status on Saturday, 07-Jan-2023 00:34:55 JST Thomas Vander Wal

     in reply to

     • Evan Prodromou

     @elipariser Do you know @evan? May be worth connecting if not. Evan had a hand (and brain) in this.

   • Embed this notice
     Evan Prodromou (evan@prodromou.pub)'s status on Saturday, 07-Jan-2023 00:36:56 JST Evan Prodromou

     in reply to

     • Thomas Vander Wal

     @vanderwal @elipariser also, we have a really dumb problem in the tech world, where we conflate scalable business models (jackpot models where small input gives great output at low probability) with scalable system architectures (can add resources to keep up with demand). It's really a ridiculous mistake, but it's also two concepts that lie very close to each other.

   • Embed this notice
     Evan Prodromou (evan@prodromou.pub)'s status on Saturday, 07-Jan-2023 00:40:38 JST Evan Prodromou

     in reply to

     • Thomas Vander Wal

     @elipariser @vanderwal what did it get you thinking about?

   • Embed this notice
     Eli Pariser (elipariser@mastodon.social)'s status on Saturday, 07-Jan-2023 00:40:39 JST Eli Pariser

     in reply to

     • Evan Prodromou
     • Thomas Vander Wal

     @evan @vanderwal To be honest I haven't been deep in the debate and thinking (as I know you have). But the discussion in this piece definitely got me thinking. https://softwaremill.com/the-architecture-of-mastodon/

   • Embed this notice
     Evan Prodromou (evan@prodromou.pub)'s status on Saturday, 07-Jan-2023 00:44:26 JST Evan Prodromou

     in reply to

     • Thomas Vander Wal

     @vanderwal I will take a look, but no, that's not the case.

     Venture investors look for "scalable" business models.

     Systems architects try to build "scalable" hardware and software systems.

     These two uses of the word "scalable" are distinct.

   • Embed this notice
     Thomas Vander Wal (vanderwal@mastodon.social)'s status on Saturday, 07-Jan-2023 00:44:27 JST Thomas Vander Wal

     in reply to

     • Evan Prodromou

     @evan it sounds like you were watching explainers I was making yesterday about basics of scaling and complexity and why the center is simple and need solid next tier that handle the deviatiations and divergent rules / constraints for all that follows in that chain.

   • Embed this notice
     Evan Prodromou (evan@prodromou.pub)'s status on Saturday, 07-Jan-2023 00:48:24 JST Evan Prodromou

     in reply to

     • Thomas Vander Wal

     @elipariser @vanderwal so, that's just not true.

     The cost of hosting a server goes up significantly when you've got superusers on your server -- accounts with many hundreds of thousands or millions of followers.

     This is true for many kinds of social networks. I know that at Twitter, it was a major problem in the early days of the company. And that was for distribution within their own environment.

   • Embed this notice
     Eli Pariser (elipariser@mastodon.social)'s status on Saturday, 07-Jan-2023 00:48:25 JST Eli Pariser

     in reply to

     • Evan Prodromou
     • Thomas Vander Wal

     @evan @vanderwal Yeah. Definitely thinking about the latter. The question I was thinking about is... if the cost to host a server goes up dramatically/exponentially as the network scales, what does that mean for decentralization?

   • Embed this notice
     Evan Prodromou (evan@prodromou.pub)'s status on Saturday, 07-Jan-2023 00:50:38 JST Evan Prodromou

     in reply to

     • Thomas Vander Wal

     @elipariser @vanderwal the obvious solution is that people who want that kind of reach should be paying for the burden they put on the network. Servers with open registration should put a limit on number of followers (~10K is a good number) and people who want to be broadcasters should host their own servers, or go to specialised servers that can manage that load.

   • Embed this notice
     Evan Prodromou (evan@prodromou.pub)'s status on Saturday, 07-Jan-2023 00:53:34 JST Evan Prodromou

     in reply to

     • Thomas Vander Wal

     @elipariser @vanderwal another issue is that with fan-out, it takes a non-trivial amount of time to deliver to each subscriber. So, instantaneous responses are hard to guarantee.

     Platforms where the publisher and subscriber are on the same server can fake this by distributing to the tiny percentage of subscribers who are online *first*. The vast majority get their messages a few minutes, hours or days later.

     We can't fake this, usually, although there are tricks we could do.

   • Embed this notice
     Evan Prodromou (evan@prodromou.pub)'s status on Saturday, 07-Jan-2023 00:57:04 JST Evan Prodromou

     in reply to

     • Thomas Vander Wal

     @elipariser @vanderwal anyway, the key assumption you made (cost per server goes up exponentially as number of servers goes up) is untrue.

   • Embed this notice
     Evan Prodromou (evan@prodromou.pub)'s status on Saturday, 07-Jan-2023 01:01:46 JST Evan Prodromou

     in reply to

     • Thomas Vander Wal

     @elipariser @vanderwal I think the big social problem is that superusers have gotten used to freeriding on the platforms.

     They provide a lot of value as content creators, and they're not used to paying for their own infrastructure on services like Facebook or Twitter.

     Ones who run their own Web sites are aware, but others aren't.

     One way might be to have special "celebrity" servers, supported by donations, where only high-follower-count Twitter or IG or YouTube users can join.

   • Embed this notice
     Evan Prodromou (evan@prodromou.pub)'s status on Saturday, 07-Jan-2023 01:07:29 JST Evan Prodromou

     in reply to

     • Shoq

     @elipariser @shoq Every post doesn't have to be pushed to every server.

   • Embed this notice
     Eli Pariser (elipariser@mastodon.social)'s status on Saturday, 07-Jan-2023 01:07:30 JST Eli Pariser

     in reply to

     • Shoq

     @shoq This is great, but I'm more talking about some of the technical/structural scalability problems... e.g. if every post interaction has to be pushed to every server, how does that hold up in a world where there are tens of thousands of servers and posts with tens of thousands of interactions?

   • Embed this notice
     Shoq (shoq@mastodon.social)'s status on Saturday, 07-Jan-2023 01:07:32 JST Shoq

     in reply to

     @elipariser You've seen this? It touches on the big hurdles to come.

     https://www.techdirt.com/2023/01/04/journalists-and-others-should-leave-twitter-heres-how-they-can-get-started/

   • Embed this notice
     Evan Prodromou (evan@prodromou.pub)'s status on Saturday, 07-Jan-2023 01:14:38 JST Evan Prodromou

     in reply to

     • Thomas Vander Wal

     @elipariser @vanderwal we can make some improvements at the protocol level by reducing the delivery overhead, say by using WebSockets to keep connections between servers alive.

     Another possibility is using lighter processes for job handlers. That's a Mastodon issue rather than a fediverse one.

   • Embed this notice
     Evan Prodromou (evan@prodromou.pub)'s status on Saturday, 07-Jan-2023 01:19:59 JST Evan Prodromou

     in reply to

     • Thomas Vander Wal

     @vanderwal I'll read through. I think a typical AP interaction has 3 hops: sender client to sender server; sender server to receiver server; receiver server to receiver client.

   • Embed this notice
     Thomas Vander Wal (vanderwal@mastodon.social)'s status on Saturday, 07-Jan-2023 01:20:00 JST Thomas Vander Wal

     in reply to

     • Evan Prodromou

     @evan Yes, agree. Just dealing with the second and explaining why keeping the center handling things as simply as possible and put the deviations to the core needs. But, also need to keep the hops away from the core to a minimum to reduce latency.

   • Embed this notice
     Evan Prodromou (evan@prodromou.pub)'s status on Saturday, 07-Jan-2023 01:26:05 JST Evan Prodromou

     in reply to

     • Ted Curran M. Ed. ?

     @tedcurran you might want to think that one all the way through. What's different between a popular Toot and a popular 2-hour movie in 4K? What are people's expectations for each?

   • Embed this notice
     Ted Curran M. Ed. ? (tedcurran@indieweb.social)'s status on Saturday, 07-Jan-2023 01:26:06 JST Ted Curran M. Ed. ?

     in reply to

     • Evan Prodromou
     • Thomas Vander Wal

     @evan @elipariser @vanderwal It seems to me that this was the problem BitTorrent was originally meant to solve - decentralizing the hosting of massively popular assets by making downloaders simultaneously upload as well. ?

   • Embed this notice
     Evan Prodromou (evan@prodromou.pub)'s status on Saturday, 07-Jan-2023 01:40:46 JST Evan Prodromou

     in reply to

     • Thomas Vander Wal

     @elipariser @vanderwal cool.

     Just remember: the trade-off is between:

     - big fan-out to lots of subscribers
     - low resources on the publishing server
     - near-instantaneous delivery

     I don't think you can have all three.

     This might already be a well-known problem. Otherwise, I claim this as Evan's Triangle.

   • Embed this notice
     Eli Pariser (elipariser@mastodon.social)'s status on Saturday, 07-Jan-2023 01:40:47 JST Eli Pariser

     in reply to

     • Evan Prodromou
     • Thomas Vander Wal

     @evan @vanderwal Thanks for taking the time to spell this out -- makes a lot of sense and I like the philosophical premise.

   • Embed this notice
     James M. (jamesmarshall@sfba.social)'s status on Saturday, 28-Jan-2023 13:36:52 JST James M.

     in reply to

     • Evan Prodromou
     • Thomas Vander Wal

     @evan @elipariser @vanderwal sorry for the necro, but regarding Evan's Triangle-- what if, for wide distribution, you use the tape tree concept? That is, the sending server distributes to a much smaller number of intermediary servers, each of which sends to a fraction of the recipient list? It would be non-trivial-- for one thing, the intermediaries would have to know which recipients they're responsible for-- but I'm not convinced it's not possible. Not much load on any one server.

   • Embed this notice
     Evan Prodromou (evan@prodromou.pub)'s status on Saturday, 28-Jan-2023 13:36:52 JST Evan Prodromou

     in reply to

     • James M.

     @jamesmarshall sure, but also not near-realtime delivery.