[pfraze]

We looked quite closely at ActivityPub. Here's why we didn't go with it:

1. AP doesn't have the facilities for global aggregation which can power search, discovery, algorithms, and metrics. The user community has been very clear that they do not want it to be introduced. We felt the connectivity of a shared global network was extremely important to the UX, but we felt it would be wrong to fly in the face of the AP world's established norms & wishes.

2. We felt that strong account portability was an extremely important feature of the system, to ensure that users don't get locked into a specific host. AP's redirection model of account migration concerned us.

3. We're concerned about the cost structure of AP. We're concerned that self-hosters are going to pay a prohibitively high price for virality. This is why we designed the network to avoid placing heavy load on PDS.

I know that the AP world is frustrated with the competition between the protocols and suspicious of how we've chosen to do things. It's a shame because I think we're after similar things, and hold similar values. We didn't set out to sabotage the AP world; we just felt like there were important changes that needed to happen for this mission to work.

Note, however: Our software is not proprietary. It's open-source. The specs are open. The network firehose is open. We're working on getting every piece of the infrastructure into good governance and straightforward self-hosting. It just takes time.

[S0y]

>AP doesn't have the facilities for global aggregation which can power search, discovery, algorithms, and metrics.

Very nice way to say "AP isn't centralized enough".

>We felt that strong account portability was an extremely important feature of the system, to ensure that users don't get locked into a specific host. AP's redirection model of account migration concerned us.

What's your current timeline to start accepting incoming account migrations back into the bluesky hosted PDS? When will account migrations officially be a recommended operation? Source: https://github.com/bluesky-social/pds/blob/main/ACCOUNT_MIGR...

>We're concerned about the cost structure of AP. We're concerned that self-hosters are going to pay a prohibitively high price for virality. This is why we designed the network to avoid placing heavy load on PDS.

First of all, self hosting an ActivityPub service is not prohibitively expensive, heck expensive just isn't even a word a would use at all. On the other hand, what's expensive is the cost of hosting the bluesky relay. What you're essentially doing is just taking on the burden/cost of data processing and hiding it from the end user. The fact that ATProto requires a relay is at complete odds with the premise of decentralization and federation. You're no more decentralized than google search giving you results from different websites.

>I know that the AP world is frustrated with the competition between the protocols and suspicious of how we've chosen to do things. It's a shame because I think we're after similar things, and hold similar values. We didn't set out to sabotage the AP world; we just felt like there were important changes that needed to happen for this mission to work.

We're frustrated with bluesky describing itself as decentralized and federated when it isn't. Look, I get it, You guys are trying to run a business. You can't control ActivityPub so you made ATProto. It's your thing so you can make what you want with it. You can make it open-source, but at the end of the day, you guys decide. Just be honest about it.

[pfraze]

Believe it or not, it’s possible to have meaningful differences about the way to design a system while maintaining the same motives. It’s clear that you’re happy with the ActivityPub design. I’m not. And your argument right now is akin to saying the Web isn’t decentralized because it uses search engines.

[S0y]

If you actually cared about having a meaningful conversation you wouldn't be tip-toeing around my arguments. There's clear dissonance between the words you're speaking and the actual reality of how ATProto/Bsky works. You say you have the same motives yet this is not what the technology shows.

>your argument right now is akin to saying the Web isn’t decentralized because it uses search engines.

Is that really all I said?

[pfraze]

Very well.

> Very nice way to say "AP isn't centralized enough".

You seem to be operating under the misconception that having large secondary indexing services in the system is the same thing as binding the system to single organizations. Anybody can run a relay or appview, same as anybody can run a PDS.

> What's your current timeline to start accepting incoming account migrations back into the bluesky hosted PDS? When will account migrations officially be a recommended operation? Source: https://github.com/bluesky-social/pds/blob/main/ACCOUNT_MIGR...

When the software is sufficiently tested and implemented.

> First of all, self hosting an ActivityPub service is not prohibitively expensive, heck expensive just isn't even a word a would use at all.

This remains true only so long as the network remains under a certain size, and your posts never go viral.

> On the other hand, what's expensive is the cost of hosting the bluesky relay. What you're essentially doing is just taking on the burden/cost of data processing and hiding it from the end user.

We're keeping the most valuable part of the system -- account ownership -- from having its costs bundled with application ops. It's important that there are hundreds of thousands of account hosts. It's only important that there are 5 to 10 microblogging applications, and that users can switch between them as they come and go.

In fact, I would argue that binding user accounts to individual application instances & their governance like AP does is a massive mistake. It's much more important that you guarantee users' continuity of identity as apps come and go.

[S0y]

I Appreciate you taking the time to properly respond.

>large secondary indexing services

We've talked about this before. The relay isn't secondary. proof of the matter is, bluesky last week went down because it was down.

>Anybody can run a relay or appview, same as anybody can run a PDS.

That's just saying anybody can fork the network if they're not happy. that's not very collaborative and social.

>When the software is sufficiently tested and implemented.

I would think this was a more pressing matter seeing your previous response.

>This remains true only so long as the network remains under a certain size, and your posts never go viral.

I think you're overestimating how taxing going "viral" is on an ActivityPub server. if one of your posts goes viral, it doesn't get hit for every follower you have. It'll only be a request per instance. Plus, task queues exist. Yes going viral is taxing on a server. it doesn't mean the solution is just to offload that burden to some centralized server.

>We're keeping the most valuable part of the system -- account ownership -- from having its costs bundled with application ops.

Except the tradeoff is relying on a handful of large organization that have the resources to burden the cost of running a network. Those networks then decide who gets to post or not. account ownership isn't worth much if you can't speak anywhere. If I were to get banned from the bsky relay, I'd be essentially barred from interacting with anyone on the ATmosphere until someone else came along and created a new relay or appview. On activitypub, maybe mastodon.social doesn't like what I say so they ban my instance. But at least I can still interact with the thousand of other instances that exist. Now you can say, don't be an ass and you won't get banned, and I agree. But when you've created a system where only large organizations have the capacity to run a network, Maybe now I get banned because Coca-Cola decided they didn't want anyone saying Pepsi tasted better on their network.

>In fact, I would argue that binding user accounts to individual application instances & their governance like AP does is a massive mistake.

I think we'll agree to disagree on this one.

As always Appreciate the debates pfraze, hopefully these conversation help users decide which platform they like the best.

[elfprince13]

> I think you're overestimating how taxing going "viral" is on an ActivityPub server. if one of your posts goes viral, it doesn't get hit for every follower you have. It'll only be a request per instance. Plus, task queues exist. Yes going viral is taxing on a server. it doesn't mean the solution is just to offload that burden to some centralized server.

I run a single-user Mastodon instance and replying to a viral post took me offline for like 24 hours.

[pfraze]

Sure. I'm really happy to debate the substance of these designs, because it is interesting and should drive decisions.

> The relay isn't secondary. proof of the matter is, bluesky last week went down because it was down.

Mmm kind of. The data is primarily stored in the PDSes and there can be a plurality of relays and appviews, none of which are considered authoritative / primary. If a relay goes down, anybody downstream of it is (fire)hosed, but that's just systems.

One useful observation -- the relay is a convenience we implemented so that work can be shared. Generally speaking an appview (or any other service) could crawl PDSes directly & sync their event streams. You're right that a ban from the bsky relay is going to affect visibility among anybody downstream of it, and that a relay monoculture would centralize control. This is why we have an organizational goal to get other independent relays running.

Regarding costs, there's a laws of physics thing at play. If you want global activity in your app, you're going to pay for it like we are. However -- if youre happy with a subset that's similar to ActivityPub, you can setup an appview which selectively syncs according to the social graphs of registered users and other known links (like URIs of replied-to posts). You could attach it to a PDS if you wanted. You then might want an additional layer for push-notifying peers for activity outside their existing social graph, though that's somewhat optional (you can get a pretty rich dataset from a pull-based crawl model). If it turned out that the global view was cost prohibitive for any org, this is the implementation I'd push for people to develop. This "mode" was in our original plans but cut for time; it's not something the protocol needs to prescribe for or against.

>I think you're overestimating how taxing going "viral" is on an ActivityPub server. if one of your posts goes viral, it doesn't get hit for every follower you have. It'll only be a request per instance.

Maybe you're right. But we wanted to target an aggressively low capital and management cost for account hosting.

>>When the software is sufficiently tested and implemented.

>I would think this was a more pressing matter seeing your previous response.

Well. We do our best to juggle priorities, but it's a small team with a lot of work on our plates.

[fsflover]

> AP doesn't have the facilities for global aggregation which can power search, discovery, algorithms, and metrics.

I guess you should also refuse to use the Internet itself, too.

> We felt that strong account portability was an extremely important feature of the system, to ensure that users don't get locked into a specific host

So how do I leave a host and join another, independent one in Bluesky?

[pfraze]

For account hosting: https://atproto.com/guides/self-hosting

For running a relay and appview, the code is dispersed among https://github.com/bluesky-social and we're working on a straight-forward distribution

[Volundr]

Mainly because your here and replying, I looked at self hosting the PDS and bounced off because there wasn't really any documentation on day 2 ops. How do I do backups/disaster recovery? What data is stored here and what happens if it's lost? What kind of traffic should I expect to see? What are the risks around updates?

I can probably figure this stuff out by learning the protocol, but I wish the documentation around hosting was deeper than "run this script to install it, run this other one to update"