The “ability” to decentralize has costs that scale quadratically. So in every practical sense, it cannot be decentralized. At best it could have a few servers that participate.
No, it doesn’t scale “quadratically”. That’s what going viral on Mastodon does to a small instance, not on bluesky. Pretty much everything scales linearly. The difference is certain components handle a larger fraction of the work (appview and relay).
Both a bluesky appview and a Mastodon instance scales by the size of the userbase which it interacts with. Mastodon likes to imagine that the userbase will always be consistent, but it isn’t. Anything viewed by a large part of the whole Mastodon network forces the host to serve the entirety of the network and all its interactions. So does a bluesky appview, in just the same way, but they acknowledge this upfront.
Meanwhile, you CAN host a bluesky PDS account host and have your traffic scale only by the rate of your users’ activity + number of relays you push these updates to. Going viral doesn’t kill your bandwidth.
I can’t speak to how traffic costs and mastodon works, but this article explains how having multiple blue skies federating with each other scales quadratically. https://dustycloud.org/blog/how-decentralized-is-bluesky/ it is very thorough.
In fact, it is worse than the storage requirements, because the message delivery requirements become quadratic at the scale of full decentralization: to send a message to one user is to send a message to all. Rather than writing one letter, a copy of that letter must be made and delivered to every person on earth
That’s written assuming the edge case of EVERYBODY running a full relay and appview, and that’s not per-node scaling cost but global scaling cost.
Because they don’t scale like that, global cost is geometric instead (for every full relay and appview, there’s one full copy with linear scaling to network activity), and each server only handles the cost for serving their own users’ activity (plus firehose/jetstream subscription & filtering for those who need it)
For Mastodon instance costs, try ask the former maintainers of https://botsin.space/
I’m sad that bots in space had to spin down, but there are still bots on Mastodon. One server quitting didn’t take everything down.
The part where if a mastodon post gets popular, it has to serve that to everyone makes sense because it’s kind of like a website. Maybe there could be a CDN like Cloudflare that a mastodon server could use to cache responses?
The part about Bluesky that doesn’t sound good to me is “to send a message to one user is to send it to all”. Wouldn’t this be crazy with even 100 servers for 10000 users, vs 2 servers with 5000 each? Not sure how the math works but it doesn’t look good if they have to duplicate so much traffic.
Nope, it’s 100% centralized.
It’s 100% centralized, but with the ability to be decentralized. Sorta like Threads before they started federating
Sure, but until it actually gets used significantly in that way, we might as well just say it’s centralized.
The “ability” to decentralize has costs that scale quadratically. So in every practical sense, it cannot be decentralized. At best it could have a few servers that participate.
No, it doesn’t scale “quadratically”. That’s what going viral on Mastodon does to a small instance, not on bluesky. Pretty much everything scales linearly. The difference is certain components handle a larger fraction of the work (appview and relay).
Both a bluesky appview and a Mastodon instance scales by the size of the userbase which it interacts with. Mastodon likes to imagine that the userbase will always be consistent, but it isn’t. Anything viewed by a large part of the whole Mastodon network forces the host to serve the entirety of the network and all its interactions. So does a bluesky appview, in just the same way, but they acknowledge this upfront.
Meanwhile, you CAN host a bluesky PDS account host and have your traffic scale only by the rate of your users’ activity + number of relays you push these updates to. Going viral doesn’t kill your bandwidth.
I can’t speak to how traffic costs and mastodon works, but this article explains how having multiple blue skies federating with each other scales quadratically. https://dustycloud.org/blog/how-decentralized-is-bluesky/ it is very thorough.
That’s written assuming the edge case of EVERYBODY running a full relay and appview, and that’s not per-node scaling cost but global scaling cost.
Because they don’t scale like that, global cost is geometric instead (for every full relay and appview, there’s one full copy with linear scaling to network activity), and each server only handles the cost for serving their own users’ activity (plus firehose/jetstream subscription & filtering for those who need it)
For Mastodon instance costs, try ask the former maintainers of https://botsin.space/
I’m sad that bots in space had to spin down, but there are still bots on Mastodon. One server quitting didn’t take everything down.
The part where if a mastodon post gets popular, it has to serve that to everyone makes sense because it’s kind of like a website. Maybe there could be a CDN like Cloudflare that a mastodon server could use to cache responses?
The part about Bluesky that doesn’t sound good to me is “to send a message to one user is to send it to all”. Wouldn’t this be crazy with even 100 servers for 10000 users, vs 2 servers with 5000 each? Not sure how the math works but it doesn’t look good if they have to duplicate so much traffic.
This is a little bit more black and white compared with the other responses. 🙈