"cheap by being shallow" - ok Christine. you invent a way to archive all of any distributed network that doesn't perpetually increase in cost. we'd absolutely love to see it over in blockchain world!
That she's still hung up on relays when the biggest centralizing force currently is the bsky appview is interesting. A non-archival relay is cheap. The bsky appview is the big archival and compute-heavy api endpoint that's hard to move away from bc the offical app points at it.
So the thing is it's actually relatively straightforward to spin up your own instance of the bsky appview. But the costs of running it scale with the number of users pretty straightforwardly and they're not, like, super low. And what that even does for you is weird bc of moderation and app defaults.
Somewhat, but more subtle and hard to explain all the ins and outs of the way the various systems interact. you have appview bans, pds takedowns, and they interact in funny ways. Like an account that was taken down and banned can be resurrected, but they're basically dark matter to the bsky appview.
being an end user that doesn't have to think about all of this except "can I shoot the shit with my friends and see the brainrot" is so much easier than obsessing over making new reasons that it's still not decentralized enough or whatever
just write your damn skeets and let the people work lol
The real risk is imho a leftist censorious mob that manages to push most users to PDSes/infra that are locked-closed à la mastodon, only websocket'ing towards relays&appviews that recursively follow their own censorious mob rules (à la https://mastodon.art)
eh that does matter to an extent but if a relay wishes to restrict access they could simply expose only jetstream. things like custom feeds have been gradually moving from the relay firehose to jetstream. would be easy to disguise a relay as a feed currently, but you can't unstrip the cryptography.
i still strongly doubt this is a risk, i think the people who host PDSes right now care too much about decentralization to play this kind of game (they'd be hosting mastodon instances otherwise) but it's not like, impossible
I mean some network path (direct or indirect) is going to be there anyways. If the intermediary is pressured to stop forwarding data (or most data), either by mob rule or by EUrocrats, that would be a problem
I can, I'm not convinced the US censorship caucus has the technical sophistication to attempt this. The EU definitely does have a surprising number of anti-speech technologists advising regulators.
Interesting: A half-connected "shadow realm" atproto network. The shadow realm nags all data from all PDSs it can reach, but only allows connections to its relays/AppView/PDS for "elite" users.
Possible, but it only would take one motivated "rogue" user to bridge that shadow network to the rest.
as others have mentioned, i’m increasingly convinced using this word is stupid and if not using the word keeps some small group of apub folks from ever using atproto meaningfully, oh well.
Apples vs oranges. We used to see this a lot in video games, the pointless debate over whether a game or dev was "indie" or not, when there were far far more important metrics to consider, like (haha) if a game was fun. 🤷
The bad news: the debate never died. Those arguing just stopped caring.
ive been wanting to write this story about how the meaning of decentralisation has evolved over time within the fediverse and atmosphere, and have spend quite some time on a draft already. but doing it justice to explain why it all got so confusing and kinda stupid is just so much work lol
feels like its kinda not worth it, so many other things to do and write as well. so if its too hard to even properly tell the story of what decentralisation even means anymore, than i do feel that your point is correct and that decentralisation in this context has lost a useful meaning
i have 2 challenges:
from what i can tell fedi used to be more open to other claims of decentralisation (nostr for example), but became more absolute towards only the email and activitypub mental model over more recent years. having a good thesis for why that is is hard
If your network topology is not identical to the ideal network topology I have pre-registered in my head, and have not shared, you are basically Facebook
- setting up a switchable client metadata oauth for local/production
- Looking at how you did the verification lookups and using constellation or the com.atproto.sync.listReposByCollection on it 👀
its interesting because a hub-and-spoke topology reduces scaling costs from quadratic to linear but quadraticity was the original critique right . i *do* think we will end up with problems with relay operation as we go from like 2k PDSes to like 20k (approaching too many open websockets!) but -
Until somebody advances the protocol so my PDS can talk to your PDS's with only DNS to guide us, it requires an intermediary and intermediary ain't decentralized.
You say that DNS is a centralized dependency, uh, um... Shut UP!!
im not even sure what this is supposed to mean, pds's can subscribe to an arbitrary number of relays, its not really a star topology from an actual design standpoint?
She’s been working on a distributed compute architecture for a long time, which is a quote-unquote truly decentralized system—essentially chasing the Holy Grail of decentralized systems. She has a vision and views any tradeoffs against her vision as unacceptable.
I like her work, particularly OCapN (https://ocapn.org/), but I hate the purity tests going on here. BlueSky /is/ meaningfully decentralized, and that’s good.
I worry about how many tech leftists want intentionally exclusive network technologies, and I think there’s some of that here too.
it's a good question!
charitably, I would say it's a narrow focus on server-to-server federation ala activitypub as the one ideal architecture for "true decentralization", whatever that means.
It is a good question... she's critiqued AP-style federation as well, for leading to a nation-state model. So maybe it's a combo of her view that Spritely's approach is inherently power-distributive in ways that others aren't, and a belief that "decentralized" means completely-power-distributive
Putting the specifics of language aside, I do think it's useful to look at power-centralizing and power-distributive aspects of the various protocols. But I *don't* think a fight over the meaning of decentralization is the best way to accomplish that!
Yeah, agree w/ this take. She remarked over on the AP side of the fence that she thought that AP relays are a bad idea because they are centralizing ¯\_(ツ)_/¯
If I get the argument correctly, it's that relays are a centralizing force because operating a full-network relay is expensive, so few are likely to exist and they become a chokepoint.
By comparison, the Mastodon "whole network" experience, or even a usable day-to-day experience, relies on your server federating with lots of other servers and people from your server following users on those servers so data is synced. There's still a "centralizing pressure" toward large servers!
Exactly! it's funny cause like… I and many others run our "independent" servers on https://Masto.host … so like… we've just pushed our single point of failure into a different service layer…
What's funny is looking at other decentralized social protocols, ActivityPub always finds itself as the outlier being made up of weird monolithic instances.
Fedi's real dislike for relays is what they originally complained about with Nostr too, since that is literally made up of a bunch of relays that you announce signed messages onto, and technically the user doesn't "own" anything.
Comments
just write your damn skeets and let the people work lol
Possible, but it only would take one motivated "rogue" user to bridge that shadow network to the rest.
Anyone can create their own walled garden network already and feed it data from outside, without giving back.
Usually this creates quite a dull place w/o much interaction.
no, i won’t allow any deviation from the model outside of what i have imagined decentralized to mean.
The bad news: the debate never died. Those arguing just stopped caring.
i have 2 challenges:
from what i can tell fedi used to be more open to other claims of decentralisation (nostr for example), but became more absolute towards only the email and activitypub mental model over more recent years. having a good thesis for why that is is hard
[a hush falls over the crowd]
Arguing about categorization is, in this specific instance, borderline useless
I’m team @hailey.at, we should simply let the word dies
You don’t need to go too deep into the details, just acknowledge that there’s now a sort of schism or whatever about it
Wild how bluesky is somehow in a constant superposition between star and fully connected topos.
Feels like @dame.is would be all over this.
Current leaderboard looks like:
1. @dame.is
2. @baileytownsend.dev
@dame.is Doesn't even know it but I dodged one earlier this week already from him already lol
- setting up a switchable client metadata oauth for local/production
- Looking at how you did the verification lookups and using constellation or the com.atproto.sync.listReposByCollection on it 👀
Until somebody advances the protocol so my PDS can talk to your PDS's with only DNS to guide us, it requires an intermediary and intermediary ain't decentralized.
You say that DNS is a centralized dependency, uh, um... Shut UP!!
I mean, she was asked for her opinion.
And each star represents an instance, like Mastodon, which everything comes down to again anyway.
That is true decentralization!
I worry about how many tech leftists want intentionally exclusive network technologies, and I think there’s some of that here too.
charitably, I would say it's a narrow focus on server-to-server federation ala activitypub as the one ideal architecture for "true decentralization", whatever that means.
https://social.coop/@cwebber/114496554706236759