Hacker News Reader

Comment by phoe-krk | Hacker News Reader

It's an error to treat the Fediverse as you would treat Twitter, where there is a singular platform in some way external to its users. Fediverse is a mesh network with single-user instances being feasible. If a single node has problems scaling, split it up. If a whole network has problems scaling, add more nodes. Fediverse is not yet at levels where the whole network is congested, either.

View on HN

This is a nice thought, but reality is always different.

A really popular person might have many millions of followers on tens or hundreds of thousands of nodes.

That person then boosts a post from another server and they have effectively DDOSed that server. All the other nodes will start querying that server about that post or image.

You can say that a single user should not have that many followers, however the root of the issue is that mastodon will have some scaling issues.

Yes, and we're already seeing it e.g. with https://mas.to getting hugged to death because of pg joining it. If anything, it can be treated as a technological problem, solvable e.g. by (theorizing now) splitting a single instance over a series of machines, or a social problem, by (not theorizing anymore) a famous enough person hosting their own instance as already suggested several times on the Fediverse. The Fediverse will eventually establish a concrete way for handling that sort of matters.

> The Fediverse will eventually establish a concrete way for handling that sort of matters.

They've had years to get this right. This is their moment to shine and they're blowing it.

> a famous enough person hosting their own instance

So if the likes Obama or Stephen King want to move over to Mastodon their barrier of entry would be to complete this laundry list [1] or find a service that does this for them? Given that they own the instance, they would also need to provide moderation for it.

I don't disagree that there are solutions to the scale problem. What I have trouble seeing are solutions that are viable. That's the thing about a centralized service--you have a company that handles all of this so your users don't have to.

[1] https://docs.joinmastodon.org/user/run-your-own/#so-you-want...

> or find a service that does this for them?

Assuming that Mastodon growth continues, such services will probably emerge (there are already a couple of Mastodon hosting services, but there's probably a market for a "so you're a celebrity who wants a single-user Mastodon instance" service which has yet to be filled).

> Given that they own the instance, they would also need to provide moderation for it.

If it was a single-user instance, that would be fairly trivial :)

I do wonder if the next phase of social media is to encourage people to use a single user instance similar to mastodon.host. It would probably have to offer more than just twitter though e.g peertube, pixelfed etc. Start with a free tier, have it all running on auto scale and then notify people when they start getting popular that they need to scale up and start paying for their instance if they want to grow their audience. Don’t know if it would work or not but it’s an idea.

Most likely not, mostly due to lack of discovery.

When you start off an an empty instance, the only thing you see is yourself. There are no trends, there are no posts, and everywhere you click is just you.

You have to start following people for it start populating, but that requires knowing strings to follow. And then you can you see is stuff in your bubble (maybe a feature?).

But it really missing the dopamine response, as after you read the stuff people you follow posted or boosted; there is nothing.

A dedicated person can push through this initial period. But for mass appeal, people will quickly bounce if they feel like no one else is there.

> That person then boosts a post from another server and they have effectively DDOSed that server. All the other nodes will start querying that server about that post or image.

But that's fixable in several ways - local media cache for an instance, the boost including the preview rather than each client fetching it for themselves, etc.

The problem* with ActivityPub is that it's been designed and developed in a bit of user vacuum and it definitely has issues but they can all be worked on. Much like Twitter going from Rails to Scala, etc., to fix their issues.

There is already a local media cache. My comment doesn't suggest that each follower requests stuff from the home server, but each instance.

The instances cache pretty aggressively, but you still have a thundering herd when a popular person boosts a message.

> There is already a local media cache.

I suppose I should have been clearer - apologies. I was talking about a cache for previews (which would presumably be folded into the local media cache since it's already there.)

> you still have a thundering herd when a popular person boosts a message

Not if the boost includes the preview - that -should- cut it down significantly (and indeed it has been proposed but is still being bikeshedded by the Mastodon devs because "what if someone sends a fake preview?!?!?!?!")

Yes, this is a real thing that will happen. However, this is no different than a popular website like, say, HN, linking to someone's hack project and giving it the hug of death. This stuff happens, and if any operational element boils down to someone's hobby then you'll see some issues. That doesn't mean Mastodon is inherently flawed any more than web pages being inherently flawed.

What will happen is the same thing that happened to web servers and email servers. Businesses will pop up to run your instance for you.

And who is supposed to do all of that? Without a central organisation doing resource planning, scaling, and paying for all of that, this is never going to run smoothly. 99.99% of users have no interest in running their own instance. Even groups or organisation will eventually get to the point where running and scaling their instance becomes a full-time job. But without any way of monetising that work, or finding sponsors, there is not much incentive to do that.

Donating to server admins seems to work for the instances whose growth I'm in/directly observing. And then, running an instance with a few hundred users is not a resource hog.

If we take the 99.99% as a hard fact that means 1 admin for 10,000 users. Sounds reasonable. Especially, if we consider that most of those will never even post anything according to Twitter behavior.

With respect to monetisation, I hope the fediverse will provide a live sandbox to try all kinds of experiments at eye level: Free, ad-supported, fees, one-time fees, pay-per-toot, pay-per-followed-instance, whatever. Multiply by private, non-profit, and for-profit approaches.

I think many people are too used to the centralised model and having one org responsible to blame for issues.

You know how we use websites right now? Who's responsible for planning, scaling, and paying for all of that? Users have no interest in running their own "website instance". Right?

And we're already at the point where for big groups running your own requires multiple full time jobs.

Even if things don't work smoothly, it's fine. Most breakage is instance local and temporary. I've been on Twitter with a weekly failwhale and it's been fine too.

This makes the network congestion exponentially worse. It’s not scalable.

Email mailing groups figured out a way of scaling despite the same sort of exponential congestion. If anything, Fediverse traffic between larger clusters may eventually ossify around some larger pipes, not unlike the Internet traffic being handled by backbone operators.

Reply on news.ycombinator.com