@kingmoth @jwildeboer @ricci I don't like federated systems where their apps (at least the ones on F-Droid) recommend a single server, it tends to cause this effect.

That being said, if a dev wants to add a server recommendation system to make it easy for new users to not burden them by having to choose a server it could choose a general server randomly from a list which would not include too popular servers (to avoid the centralisation issue) nor too small servers. Well, what if server owners could opt in such a list? 

@Commander1024 what's your source for that? https://fedidb.com/stats suggests Hetzner is responsible for about 9.3% of the fediverse.

I would say AUTOMATTIC is overrepresented because Wordpress.com blogs aren't as important for the fediverse as the number of them would suggest but still it would be nowhere near 70% even if you just removed AUTOMATTIC. DigitalOcean seems to be ahead of Hetzner too.

@Commander1024 @jwildeboer @ricci

Hetzner has 9% of all instances and 36% of all accounts.

ref: https://fedi.wrm.sr/

@Flo_Rian @jwildeboer That's an interesting suggestion, and I'm divided on it. On the one hand, I don't think it's 100% clear that the breakpoints used in antitrust are necessarily the right ones to use in this context, so staying fuzzy on the actual gauge might avoid suggesting sharp cutoffs where they don't really exist. (I did feel the need to include them in the text to give *some* context.)

On the other hand, I do think it would be good to show that there is still a benchmark for the fediverse to hit, that it can still get better

@jwildeboer
ref:

https://framagit.org/hubzilla/core/-/commit/15e836b7dd2fed3f97f134928d88e13db1d8a7f4

https://bsky.social/about/blog/02-22-2024-open-social-web

@jwildeboer I think it's important to note that this debate has alienated the entire ActivityPub community from one of the most successful digital communities for marginalised people we've seen in quite some time.

Blacksky is accomplishing tremendous things on AT, and AP has a lot to learn from that endeavour.

Rudy deserves better from the ActivityPub community.

https://www.blackskyweb.xyz/#01-read-me

@iju @jwildeboer

This is complicated.

What I'm counting is the PDSes that the bluesky relay reports knowing about, and how many repos (user data) they report having.

The number this returns is different from what you may see in other places. You'll see 38M for the number of total accounts. You'll see a few million listed for Monthly Active Users. My understanding is that what the relay reports is *some* measure of active users (it's the number of accounts the relay is currently maintaining data for) but it's not exactly MAU.

The Fediverse numbers are MAU, but it's also notorious that you can't get a fully accurate number for this because not all instances choose to opt in to such data collection, often by misreporting numbers.

@xerz @jwildeboer Oops! Thanks for pointing this out!

Fixed

@irelephant @jwildeboer In a recent interview with Rudy of blacksky, he described the PDS as the most complicated part (and they should know, they have re-implemented or forked basically all the parts now). Yeah this surprised me too.

The anecdotal information I've found on the appview itself suggests that it follows a similar distribution to the PDS, but with far fewer of them so far. I'm looking for a data source, though.

Above the appview layer, I do think feeds and labelers are likely to show interesting decentralization; labelers are probably possible to get actual usage data on, but AFAICT only bluesky can provide data about the use of feeds, and they don't (all that's available is the number of likes, which is not the same thing.)

@gabboman @jwildeboer

There is a difference between the protocol enabling decentralization and decentralization actually happening in practice. atproto is designed to be decentralized (well, with the exception of did:plc); the question of course naturally arises whether it is getting used that way or not. Clearly, *at the level of where user data is stored*, that is not currently how the vast majority of its deployment has been done in practice. This very well could change, and the protocol is (again, with the exception of did:plc) designed to facilitate that change. Will it change? I don't know, let's watch and see; that's what the site is for.

But I also want to say that being new and big is actually a potential impediment to atproto becoming more decentralized. Most systems *tend* to trend towards more centralization, and network effects *tend* to keep people in the places where lots of other people are. So, while atproto is *designed* to permit decentralization that does not guarantee that it will *happen*, and starting as one big ecosystem controlled by a single entity makes it *less likely* that it will achieve meaningful decentralization. There are people trying, though - blacksky being the clearest example. Is this the start of an inflection point where more communities spring up and migrate off the centralized infrastructure? Maybe! Maybe not! Let's watch - that's what the site is for.

@hadronized @jwildeboer In this context, the question I'm asking is whether the *ecosystem* is decentralized with respect to where data is stored.

So, in this case, sourcehut is one participant, as are github, codeberg, gltlab, all the forgejo instances that my data source (Software Heritage) knows about. The question being looked at here is whether user data (in the git world, repos) are concentrated on a few participants or spread out across many.