What happens when you have your own Mastodon instance (just for you) where over 21,000 people follow you and when youโ€™re following over 4,300 people?

You end up paying ~โ‚ฌ50/month for Mastodon hosting ๐Ÿ‘€

It also opens up interesting questions: what happens when a popular account joins your instance (hint: it will probably cost the instance maintainers quite a bitโ€ฆ I donโ€™t envy the mastodon.nu folks right now).

#fediverse #mastodon

This stuff is never free anywhere but the scale of Big Tech insulates you from it somewhat.

I actually think itโ€™s good to be reminded that our ability to communicate comes with a cost โ€“ in terms of resources, environmental impact, etc., not just money.

But, equally, it also makes the case that a system optimised to host hundreds of thousands of people on a single instance is not also somehow magically optimised to host just one person.

The latter is the problem Iโ€™m exploring with #SmallWeb.

@aral Big Tech did not insulate us from paying for services, they used our personal data to pay for it. Big Tech also made it so that nobody needed fiber optics because they did all the processing.

@gert @aral An interesting question. What would a more distributed framework look like? What would it require?

For example, to make something like Youtube without that centralization would be *extremely* difficult, it demands hierarchical caching.

@Elucidating @aral I think Peertube is a hybrid using both AP and P2P protocols. It's all about protocols.

@gert @aral I agree about the protocols bit. I managed some of the infrastructure and I work with some of the other infrastructure that makes Youtube. The amount of bandwidth required to do youtube (or netflix if we go back a few years) would crush the internet. P2P increases this problem dramatically.

Peertube is cool for what it is, but it fundamentally cannot handle a substantial fraction of YT's volume. We'd need to leverage caching more.

@gert @aral I can talk more about what that looks like if you don't know and would like to. It's not impossible. But I don't think folks have really considered things like "open source, verifiable, community lead network caching infrastructure" very much. It's an interesting space few people really understand the impact of.
@Elucidating @aral You mean something like IPFS run by communities? Sounds good to me. ๐Ÿ˜ƒ
@gert @aral Perhaps, but we'd need to find something more fit-for-purpose than IPFS.
@Elucidating @aral I'm all ears..

@gert @aral IDK if it exists. What makes youtube work is a series of expensive hierarchical caches distributed globally. There are parts of the world where this network is thin and youtube is really a suboptimal experience.

I dunno if folks really realize how much geography affects their internet. Odds are, most Americans have participating caches peered with their provider's hardware.

@gert @aral The main challenge is finding a way to distribute hot datasets to people in a way that doesn't run back to backbones and choke the life out of them. People don't realize, but a lot of the money and capital of various streaming services and content providers goes into caching infrastructure.
@Elucidating @aral Which they probably call CDN. IPFS is more like a shared filesystem..
@gert @aral Yes, we typically call this a "CDN" although I think that it's a bit imprecise. Some folks actually have network assets backing it, whereas others just have machines globally that can sync+serve.
@Elucidating @aral They might even use rsync internally.. I understand that that type of "caching" works for them, but it is not peering.
@Elucidating @gert @aral is there any more information on YouTube's caching? Do they trust ISPs with certain https certificates?

@supersingular @gert @aral I'm not sure what's been publicly shared, so I can't say more. As my job at Google works directly with these in some cases (and for awhile I was oncall for something similar to them but not quite the same), I know quite a lot more than the public does.

You can infer a lot about how SSL certificates are managed by media CDNs by examining the certificates, if you're curious. That info is public by necessity.

@gert @aral More specifically advertisers are paying for ads that are personalized via the use of our data. As much as it's not ideal, could you imagine how much youtube would cost if it wasn't for advertising: bandwidth, storage, encoding, paying creators in addition to maintaining the site. Youtube must be the single most expensive online service ever, and they're still managing to stay afloat.
@aral

> But, equally, it also makes the case that a system optimised to host hundreds of thousands of people on a single instance is not also somehow magically optimised to host just one person.

IDK; one of the major backends on the #Fediverse is #Pleroma and its cousins, which despite being more efficient (thus scaling better) than #Mastodon, it's largely used for single-instance users.
@aral on another note, how much would self hosting mitigate these costs?
@realcaseyrollins Iโ€™m assuming it would somewhat if you donโ€™t count the cost of your time :)

@realcaseyrollins Yeah, as far as I know, Pleroma is much more tuned for the small instance/single-tenant use cases.

Although, ActivityPub itself, of course, isnโ€™t.

@aral maybe some sort of peering?

Itโ€™s possible if all the celebs join one instance the cost will have a plateau, because Justin Bieber and Selena Gomez are followed by people on the same servers.

@aral any indication how frequent @mastohost clears out the media cache? I have the little moon plan and Iโ€™m already near 8gb of media. To move up to the next tier isnโ€™t a big dealโ€ฆbut knowing if/when the media cache is cleared out wouldnโ€™t hurt โ€” especially if that will let me not bump up to my limit in a week or two.

@chris @aral

- 7 days for media files in posts
- 30 days for preview cards in links
- 90 days for preview cards in other media types (videos/photos)

Source: https://masto.host/remote-media-cache-backups/

Remote Media Cache Backups | Masto.host

Masto.host was built from the ground up to make running a Mastodon instance easy.

Masto.host
@mastohost thanks for the clarification. Now that I see that I read it before, hopefully I stay within the storage I have now! ๐Ÿ˜ฌ
@chris @mastohost Wouldn't it just bump out your oldest cache items if you do fill it up?
@chris @aral @mastohost If they're using the default cache settings, then it will be weekly, but it might be worth contacting them to find out since it doesn't look like they have that information on their site.

@aral Great food for thought regarding Fediverse economics. I think it makes for a compelling argument to investigate Mastodon alternatives.

Iโ€™m considering starting a Pleroma instance. It seems to be a lighter weight option, and for whatever feature parity it misses with Mastodon, might make up for that fact being compatible with Mastodon client apps. https://pleroma.social/

Pleroma โ€” a lightweight fediverse server

@leagrr @aral I'm currently running Pleroma. Evidently it's not under active maintenance, I accepted this risk for less money out of pocket.

My Instance is a single core VPS with 2 GB ram and works without issues.

Not all apps work worked. Tusky is the one I'm currently using and works without issues.
@aral @boynux Ah, thanks for the testimonial. I am interested in running a similarly lean instance and seeing how far we get before load interferes with functionality. Curious if youโ€™d be willing to provide any other benchmarking data (number of users, task rate, etc)?
@boynux Also curious which apps seem incompatible. Thanks for indulging me, Iโ€™m fascinated.
These are the ones I tried so far:

Tusky: Just works
Fedilab (paied): Just works
Mastodon app: It took a few tried to sign in
Tooot: Didnt work
A few more data points from my notes:

Mastodon app: Search crashes the app, also sometimes hows "Not implemented" error, which indicates a missing API I guess
Fadilab: Following users works but the profile state doesn't change.

I guess so far the best one is still Tusky (Although I like tooot more)
@boynux I still need to try Toot. Thanks again for sharing your experience with Pleroma.
Always, let me if I can help and happy hacking โŒจ๏ธ

@aral this probably mainly demonstrate how resource-heavy Mastodon isโ€ฆ I would be curious to know the result in a lighter software such as Pleroma (which is very similar, feature-wise).

And this is an issue for small hosting and big ones tooโ€ฆ

@Lapineige @aral
Not quite so big numbers, but I am operating several Newsbots with approx. 1000 Followers on a single core 1 EUR VPS with 512 MB RAM with honk https://humungus.tedunangst.com/r/honk
@Haydar which at least proves it's possible to run a similar service with very few ressources.
@aral
@Lapineige @aral Yesterday I set up a mastodon instance on a USD 7/mo Digital Ocean droplet. It crashed repeatedly because it used over 100% of CPU and it wasnโ€™t even federated yet! I had to go up to the 14/mo droplet to get it to run functionally with only me, no followers and unfederated.
@katebranden @Lapineige @aral for that price you get usable machines at ovh, scaleway, ikoula or hertzner
@katebranden @Lapineige @aral I might be looking at $50/month on a droplet just to support Masto for my familyโ€™s private accounts ๐Ÿ™ƒ
@aral Excited to see where you go with this. I'm dubious about the benefit of moving from corporately-controlled platforms to individually-controlled fiefdoms. Full decentralization is what we need, not cyber-HOAs.
@aral You can find much better deals by running your own VPS. I have a 3CPU/2.5GB/40GBSSD/6.5TB-bandwidth-er-month VPS running my instance and it costs me US$24 per *year*. See https://lowendbox.com/blog/2-usd-vps-cheap-vps-under-2-month/
2 USD VPS โ€“ Cheap VPS Under $2/Month (Updated September 2025)

Are you looking for a VPS under $2 USD per month? If youโ€™re new to LowEndBox โ€“ welcome! LowEndBox.com is a blog site made up of a small team of passionate individuals that have a strong love for the web hosting community with a focus on featuring cheap and affordable VPS deals. Our mission is [โ€ฆ]

LowEndBox
@bashley @aral Your VPS has 2.5GB of RAM, in his screenshot you could see that Sidekiq alone uses almost 20GB of RAM. I guess that would also make the VPS a tad more expensive ๐Ÿ˜‰
@mediafinger @aral Ouch, that queue is pretty backlogged for sure! I suspect that sidekiq is like other services that expand to fill a % of available RAM and then settle - at any rate 20GB on only 4 threads(mine is 427 MB on 25 threads) sounds more like a memory leak to me, maybe slowing things down with swapping.
@bashley @aral wow this is amazing, how is the performance?
@boynux @aral on mine the queues don't grow, response is snappy, load average usually under 0.25, 91MB mem free.

@aral I've just today written a blogpost on this topic. Well, not the costs, but the fact that the fediverse is inherently inefficient. And why this isn't a problem (other than higher overall monetary costs)

https://berk.es/2022/11/08/fediverse-inefficiencies/

The Fediverse is Inefficient (but that's a good trade-off)

@aral very interesting! I just moved to my self hosted instance, now I'm thinking I should try not to get as popular as you ๐Ÿ˜…

@aral A single user instance may have trouble just getting a patreon from it's users too. ๐Ÿ˜†

Is it mostly the people following you or the people you are following I wonder? Easier to cut the people you follow I suppose.

@aral interesting! I admit that although I had always worked with the web, I still have this โ€œit just cloudโ€ mentality
@aral I really like the perspective here. It's not all that different than using someone else's file storage/email vs your own. There is always a cost it's either your money or you as the product.
@aral Just here to say that I agree strongly. Would be good if the lie of "free services" would come to an end.
@aral One of my first thoughts was to find out how much my use of the system costs the maintainers, then kick back 7-10x of that to do more than just defray costs. Not everyone is in a position to pay, plus turnover, expansion, and resiliency isn't free but nor is it easily factored into costs. I don't know how a system is sustainable without being up front about its costs and needs.
@aral pleroma is better optimized in that regards
@aral What sort of storage demands are you making?

@gruff Iโ€™m mostly like โ€œyou better store it, or else!โ€

(Sorry, I donโ€™t get the question. Apparently Iโ€™m using > 20GB of database space โ€“ whatโ€™s taking all that up, I donโ€™t know โ€“ and the masto.host interface shows 0GB of media storage being used, which Iโ€™m assuming is a bug.)

@aral I was just trying to get a handle on where the costs were for your plan. Media storage would normally be the killer, I would have thought, if hosting stuff for others. But if it's just 'busy' because of all your users I guess there's lots of network traffic.

I just thought it a lot of money given the costs of a small OVH VPS self host sort of approach which is less than ยฃ5 a month (40Gb storage and unlimited traffic).

@aral The .nu server must push out each post by [redacted] to almost every instance's inbox by now. That's some thousands of calls with payload. And then handle all the callbacks and validations.
But also handle all incoming requests for each mention and reply. It's a hard time for this server, that's for sure.

What could be done to improve this? Batching? Delaying? E.g. like SMTPs error 451 (not available now, try again later). More proxies?

I guess the fediverse needs to sort this out fast