There are many types of Bluesky servers. This post is about ATProto Relays, which consolidate all the data across the network into a single location. They are necessary for efficiency, but they are extremely expensive to run. Currently, there is only one ATProto relay, but there is an initiative to launch a third-party relay.
In addition, it’s estimated to current cost about $500 a month just to store the contents of a relay server. this doesn’t include network or computer cost. this will only get more expensive and lead to big businesses being the only players. relays, being the “Post office” of bluesky, have the power to suppress whatever they don’t like.
In terms of infrastructure, Bluesky does seem to be capable with keeping up with the current level of 5,000,000 posts per day by pouring money into their relay. What I’d like to know is the cost per user of Bluesky compared to a centralized service like Threads.
I think one of the bigger issues with Bluesky scaling is moderation. Mastodon admins only have to deal with moderating their own instance, while ATProto labeling services can receive reports from any and all Bluesky users. So far, most people who attempted to launch an independent replacement for Bluesky Moderation Service have failed to keep up with the volume of incoming reports. Usually, they have to narrow their scope and focus on very specific issues, like Laelaps, Asuka’s Anti-Transphobia Field, Blacksky Moderation, and AI Mod. However, as Bluesky grows, I even expect them to be unable to keep up.
Basically, on ATProto, everyone stores their own data on a PDS (Personal Data Server). This includes every post, image, video, follow, like, block, etc. The relay crawls the open web for PDSs to consolidate into a single stream of data.
From here, services can build off of that single data stream without needing to do any crawling of their own.
Of course, this does give the operator of the relay a lot of power. If they were to block your PDS, then only services relying on a different relay can access your data. This is especially why it is important that there exist independent third-party relays, but no one has taken up the mantle yet.
The PDS design is a nice approach to making AT Proto distributed. But, the way they chose to operate relays means that the protocol requires massive investment to run.
The absolute minimum spend for a node right now, even with Bluesky still growing, is something like tens of thousands of dollars per year. If Bluesky does become the next Twitter, it will probably be $100k+ per year to run a relay. That means a company could run a relay, maybe a university could run one, but it’s way out of the reach of anyone but the richest of individual users.
Can’t someone theoretically make an Activitypub relay which hosts information from all private instances that federate with it? Like, how is this (AT Protocol) better
Is this really how it works? Is that per year? Hosted? Or do you own the software and can host it on your own?
There are many types of Bluesky servers. This post is about ATProto Relays, which consolidate all the data across the network into a single location. They are necessary for efficiency, but they are extremely expensive to run. Currently, there is only one ATProto relay, but there is an initiative to launch a third-party relay.
In addition, it’s estimated to current cost about $500 a month just to store the contents of a relay server. this doesn’t include network or computer cost. this will only get more expensive and lead to big businesses being the only players. relays, being the “Post office” of bluesky, have the power to suppress whatever they don’t like.
ActivityPub really is “for the people”.
But can ActivityPub scale?
Can AT Proto? It goes both ways.
In terms of infrastructure, Bluesky does seem to be capable with keeping up with the current level of 5,000,000 posts per day by pouring money into their relay. What I’d like to know is the cost per user of Bluesky compared to a centralized service like Threads.
I think one of the bigger issues with Bluesky scaling is moderation. Mastodon admins only have to deal with moderating their own instance, while ATProto labeling services can receive reports from any and all Bluesky users. So far, most people who attempted to launch an independent replacement for Bluesky Moderation Service have failed to keep up with the volume of incoming reports. Usually, they have to narrow their scope and focus on very specific issues, like Laelaps, Asuka’s Anti-Transphobia Field, Blacksky Moderation, and AI Mod. However, as Bluesky grows, I even expect them to be unable to keep up.
How? And the fediverse works the other way around.
Basically, on ATProto, everyone stores their own data on a PDS (Personal Data Server). This includes every post, image, video, follow, like, block, etc. The relay crawls the open web for PDSs to consolidate into a single stream of data.
From here, services can build off of that single data stream without needing to do any crawling of their own.
Of course, this does give the operator of the relay a lot of power. If they were to block your PDS, then only services relying on a different relay can access your data. This is especially why it is important that there exist independent third-party relays, but no one has taken up the mantle yet.
The PDS design is a nice approach to making AT Proto distributed. But, the way they chose to operate relays means that the protocol requires massive investment to run.
The absolute minimum spend for a node right now, even with Bluesky still growing, is something like tens of thousands of dollars per year. If Bluesky does become the next Twitter, it will probably be $100k+ per year to run a relay. That means a company could run a relay, maybe a university could run one, but it’s way out of the reach of anyone but the richest of individual users.
Can’t someone theoretically make an Activitypub relay which hosts information from all private instances that federate with it? Like, how is this (AT Protocol) better
They already exist, they range from smaller webring like relays to ones that tie the major nodes together.
Neat! Does that mean there’s an ActivityPub equivalent to the ATProto Firehose?