Email or username:

Password:

Forgot your password?
Top-level
Chris Trottier

People ask if the Fediverse can scale. They wonder if decentralized social media has the wherewithal to handle "serious" network traffic.

1 billion messages per month should be the answer!

How do we do it?

It's quite simple: horizontal scale is scale.

16 comments
KarlE

@atomicpoet with all respect and gratitude, the #Fediverse should not be content with how it works so far. It scales by omission - in a way what they accused twitter, it just happens for different reasons.
When three people on different servers click on a post, they might assume they get to see the full thread. However, they will likely see three different selections of responses, one may be lucky and see them all because he shares the server with the original poster, 🧵

KarlE

@atomicpoet one may see a big proportion because they are on a large server where someone follows someone else and it gets propagated. a third will think why the heck does this not get any responses because they only follow the original poster and nobody else on their server is interested in the topic.
Subscribe to a hashtag and what you get will differ immensely.
I do not think this is a situation that is desirable and we should be complacent about.

KarlE

@atomicpoet I think when clicking on a post in the web "client", my server should fetch the thread. This would allow meaningful discussion on a common ground.

Simon Lucy

@atomicpoet

That's not exactly scaling. If a single instance fails completely then all its users fail to get service, can't even login, that isn't scaling it's a Point of Failure. That instance could itself be scaled but it can't hand off the load to other instances and users can't migrate without their instance being up.

There are other distributed but non scaled features like attachments, in that case attachments are a burden for all instances and probably a limiting factor.

Chris Trottier

@simon_lucy And what if a centralized service goes down? Well, you’re shit out of luck. Which was the case with Google+, Friendster, Vine and a whole lot of services.

On the Fediverse, the network persists even if one instance goes down.

I can also still read messages from instances that are no longer even active.

Simon Lucy

@atomicpoet not if it's scaled. Yes if it ceases to operate at all. Just like Mastodon instances that fail. You may get historical posts on Mastodon because they are on other instances but that is not scaling that's a kind of eventual consistency.

Don't confuse scaling for traffic with resilience of an overall service made up of independent operators. Resilience, product resilience is important but it is not the same as scaling to a level of traffic that's magnitudes greater than at present.

Chris Trottier replied to Simon

@simon_lucy You miss the point. Scale doesn’t even matter if Big Social pulls the plug on the service.

On the Fediverse, that will never happen.

Whatever scaling benefits you believe come from centralization ultimately means jack if the service lacks persistence. Which eventually is going to happen for Big Social—it’s always a matter of time.

Meanwhile, the Fediverse is now sending 1 billion messages per month. And it’s doing this because 22,000+ instances are federating right now.

Simon Lucy replied to Chris

@atomicpoet

At which point did I say centralisation?
You're using the term scaling in an entirely inappropriate way.
How many of those billion messages are unique? How many get to be viewed or themselves cause further messages. Just spouting numbers out of context makes no sense.

Scale always matters and it makes no difference whether it's what you call centralised, single ownership, or multiple individual ownership.

Chris Trottier replied to Simon

@simon_lucy Bullshit. I'm using scaling in the context of "How does 1 billion messages get sent."

And it's 100% true that the Fediverse is capable of sending 1 billion messages per month, as we've already just seen.

mekka okereke :verified:

@atomicpoet

I'm less concerned about Mastodon itself "scaling" and more concerned about it DDOS'ing tiny sites all over the internet. This problem is still getting worse.

Mastodon encourages a resurgence of blogging, but simultaneously can increase the technical burden of hosting a small blog.

I'm also not that worried about ActivityPub "inefficiency," because at some point, some working group is going to finish a spec for an updated ActivityPub, and servers are all going to slowly migrate.

Chris Trottier

@mekkaokereke Now this is quite a valid concern, and something that has frustrated me at length.

I actually do wonder what will happen once there’s 100 million active users, and a site goes viral.

Enrique Barcelli

@atomicpoet @mekkaokereke

This is a good point. Each site fetches the posts from the original site, right? so the only 'efficiency' comes from multiple users in one instance sharing the same fetch.

There are about 23,000 instances at the moment and about 10,000 users.

Is it possible to model/simulate some scenarios for something going viral and the kind of burden they may put on different types of instances? 🤔

Tony Hoyle

@atomicpoet
If it's simply boosted a lot, not a lot.. the post gets spread around servers and the original site doesn't see much of a hit.

If it's a link to a site, getting buried is a risk that has existed forever.. partly why things like blogspot took off.
@mekkaokereke

Daniel Freysinger

@atomicpoet Speaking from an American perspective, there are many who don't believe anything can succeed without a profit motive. Some of the best things in life are free.

Industrial Gender Accident

@atomicpoet @soatok also, the quality of interactions with others tends to be far better than on twitter. It reminds me of the way twitter used to be a decade or more ago.

Go Up