Content
@
0 reply
20 recasts
20 reactions
boscolo.eth
@boscolo.eth
One thing I like about the Bluesky architecture that I would love to see Farcaster adopt is the concept of a PDS (Personal Data Server). This way every post I make is stored in a server I control with my keys (but not necessarily host). It would help move us toward this idea of Publish (on your) Own Site, Syndicate Elsewhere: https://news.ycombinator.com/item?id=29115696
4 replies
11 recasts
59 reactions
horsefacts
@horsefacts.eth
Farcaster had a very similar design before we introduced Hubs. The problem is that if you want to get the latest state of the network, someone needs to coordinate retrieving the latest updates from many different individual servers (which fail, go offline, slow down, etc.) Bluesky uses a central "relay" that watches all the PDS servers for updates and converts them to a single stream. Theoretically anyone can run a relay but in practice it's very expensive. So it's a different trade-off. We want everyone to have access to global state without a relay. IMO the design of FC pushes the tendency to centralize further out to the edges, into client like Warpcast instead of infra like the relay.
2 replies
0 recast
10 reactions
shazow
@shazow.eth
When you say "access to global state", what state are you referring to? If we're building something like Drakula, should we also have to sync every other cast in the network too? Maybe there's a model where each atproto-style lexicon namespace is effectively a rollup that can be sync'd separately?
1 reply
0 recast
1 reaction
horsefacts
@horsefacts.eth
I'm referring to "all the casts." Namespaced rollups are a cool idea. Eventually we will need to shard the global state. (Snapchain is designed with this in mind). A challenge with sharding is that it kinda degrades to something like federation if you don't choose an unopinionated scheme. Sharding by app rather than by some naive scheme like FID makes it easier to defederate all the Drakula users at once.
1 reply
0 recast
1 reaction
