@fediversenews
So poking around fedidb.org, I thought I'd do some back-of-napkin analysis (ie procrastination).
* 50% of #fediverse users are on the largest 20 instances
* ~50% of MAU are on the largest 28
* 77% users are on instances with >10k users (largest 130)
* 73% MAUs on >10k instances
* Of largest 20 instances, user growth/mnth (trend over prev 3mnths) is ~1-2% ... with 2 major outliers:
- misskey.io: 30% (See @atomicpoet to follow the *key story)
- #mastodon.social(!!): 5%
/1
Big surprise for me was how much mastodon.social is growing (especially given the tension about its dominance).
It's current trend is to grow by the equivalent of the 23rd largest instance per month while most (ie largest) instances are growing by 2% per month (ie a few thousand, which, guessing, would be something like the 300th largest instance or so ... ie small).
/2
Also, depending on what counts as "large" (>10k users?), 70-80% of fediverse is probably on large instances, which must have implications for federating/moderating with safety in mind.
If you were admining an instance that cares about safety, when do you start with the "federate" list rather than "defederate" list and ignore all the big ones (>20k?)
Anyone doing this sort of analysis properly? Could #fedidb include such?
/end
An additional observation on MAU to User ratios.
Of 150 largest instances (min 8k users, extent of my "analysis"), the MAU/Total User ratio had no relationship with the total user count. Said ratio varies from 0-40% with outliers 48, 53, 75%.
Though, all high MAU ratio instances are small.
That is: all >30% MAU ratio instances have ~30k or fewer users. So small instance community is probably working well, but not reliably so?
Fedidb records ~21k instances.
Which means that ~20% of the fediverse users not on the core/central 150 instances are spread out amongst ~20,000 instances.
I'm guessing many of these are self-hosting, but it'd be very interesting to look through all of the data to get a clear picture.
Otherwise, there are probably many small community instances out there ... which is awesome!
@maegul @fediversenews Hi from one of those small instances: 9 AU.
instances.social sees about 17k instances. These account for ~25 million statuses/month - far short of the billion @Gagron and others have posted. With ~1.6 million AU (av 660/month each) this is implausible.
Maybe the ~4k extra instances you see may comprise the largely de-federated including prolific spambots pumping out nearly a billion which go nowhere near any instance with a decent blocklist.
Do you have a view?
@stuart universally blocked and spam-bot instances make a lot of sense.
Otherwise, I’m just relying entirely on the data from fedidb.org.
@maegul @fediversenews I'm having a bit of difficulty replicating your math here:
infosec.exchange 19,510/53,408 = 36.5%
hachyderm.io 16,098/46,990 = 34.3%
troet.cafe 12,748/39,331 = 32.4%
If you're categorizing these as small instances based on their MAUs (instead of overall users), then by my count there are only 6 such "non-small" instances.
@pevohr @fediversenews Fair. I was trying to give an impression of the shape of the data. From memory, these are the outliers in terms of size. Generally, there’s a clear pattern that higher MAU ratio instances are on the smaller end.
I don’t follow what you mean by 6 non-small instances?
@maegul @fediversenews Just that fedidb only lists 6 instances with >30K MAU, as opposed to 39 with >30k total users.
I only skimmed down to the >20k cutoff, but agree that many of the older/larger instances tended to have lower MAUs. Those three outliers definitely outshine their similarly-sized peers. (Unclear why they're doing a better job at staving off churn.)
@pevohr @fediversenews well two of them are tech focused so I figure that has something to do with it as this place seems to attract and work for tech people.
@maegul Definitely a strong hypothesis.
It's also possible that so far the current approach to moderation on those instances has been a good match (scale-wise) to the homogeneity of their respective communities.
@pevohr yea, and strong professional or hobby or lifestyle alignment between instances and its users
@maegul @fediversenews I don't have the required expertise in statistics, but I think that large instances are smaller than you think, maybe > 3K users. Tiny ones are < 10-15 users.
From instances.social, today: range of # of users, versus # of instances. For example, 234 instances have 400 to 1000 users. No MAU stats, sorry.
1-3 9447
4-9 3043
10 |- 40 1924
40 |-100 687
100 |- 400 622
400 |- 1K 234
1K |- 4K 278
4K |- 10K 82
10K |- 40K 79
40K |- 100K 17
100K |- 400K 9
400K |- 1M 1
1M+ 1
@jcastroarnaud @fediversenews
Awesome! I was hoping someone would have these numbers. So about half of instances are single user?
Actually, I think what counts as large isn’t merely statistical, but functional. Once an instance has too many people it can be harder to moderate and more suspicious when it comes to federation, because even 0.1% of users can be dodgy and ruin it. I allude to this in my comment about safety. What’s “too many”? I guessed 10,000. But it’s complex obviously.
@maegul @fediversenews More like a third of the instances, instead of half: 5069 of 16424 instances have 1 user, 2812 have 2, 1566 have 3, 991 have 4, 12490 have less than 10.
And I agree: things are more complicated than just number of users. My guess is that the moderators' work is proportional, not only to the # of users of the instance, but also to the # of users, of other instances, visible from the instance.
I would guess that instances 2 or 3 users could very well be single-user instances with a couple of accounts for various purposes, but point taken.
There's probably an interesting issue around how the number of moderators scale with the size of an instance and how it needs to scale for effective moderation. Moreover, this scaling may very well change as the fediverse grows and more mainstream users come over.
@jcastroarnaud @maegul @fediversenews
/me critiques: "accounts ≠ users"
@AlisonW @maegul @fediversenews That's fair. I'm being sloppy, *and* instances.social uses the name "users" instead of "accounts".
Unfortunately, I didn't find easy-to-collect MAU stats on instances.social, so I can't estimate actual users, instead of accounts sitting idle.
I wonder if the percentage of active users on a instance is related to the instance's size...
@jcastroarnaud @maegul @fediversenews
It's fine, nearly all services seem to make the error that people may have multiple accounts, probably because you can't programmatically know the actual number.
Small instances are more likely to be personal so more used or not used at all, I agree .
@jcastroarnaud @AlisonW @fediversenews I'm pretty sure I mentioned that in my thread somewhere ... I graphed the MAU Ratio to Instance size and there isn't a relationship at all it seems (the linear regression had R^2 of basically 0). Apart from the point I mentioned about high MAU ratio instances tending to be smaller.