Most clients can’t open a couple hundred websockets at once. Browsers are even hard capped.
So yes - it works ok with the current number of relays and distribution of users across them. But if we actually succeed in growing nostr and encouraging users not to be on the same handful of relays, then what?
No one is saying we should "encourage users not to be on the same handful of relays". I at least think the ratio of relays/users will *decrease* because your point about performance is directionally correct. The outbox model still needs to work though, because *if* someone is on a personal relay, they should still be discoverable with little to no penalty.
People are absolutely saying we should randomize relay starting lists and start users on “small” relays so
I’m not sure why you aren’t hearing/seeing that - it’s pretty common rhetoric on my feed.
I certainly agree that we need a scalable solution to finding user events in a (reasonably) censorship-resistant fashion.
Yeah, I see it too, but I don't think most of those people know what they're talking about. The four of us who went on CD last week seemed pretty well in agreement about that. I do think some randomization is good, among the top 30 or so relays (Primal is adding this, and it's good). But that's not the same as true randomization across the whole set of available relays, which would be an awful idea for several reasons.
So…can we agree some people are talking about randomizing relay lists now?
nostr:note17qa7as9y4eqksln3rdsmes74y7lv9ua5hcl295cd89sk59dz7h5svza9eh
Haha yes ok you win. I thought Primal was going to be more conservative and base selection on on some quantifiable relay characteristics. A floor of 100 relays is definitely higher than it needs to be, and will put stress on outbox-enabled clients. I think we can still handle it by rotating through connections, but it definitely will be interesting.
> 10-20 huge hubs and a long tail of 10k relays
Ironically this sounds pretty similar to the Bluesky/AT Protocol model where the indexers are the hubs and the instances are the relays.
Is it? I thought AT replicated everything to all instances. That's definitely not what I have in mind here.
There are no instances in AT, there is just one big server ran by the Bluesky Corporation. They used to call it "BGS" (Big Graph Server) but now they call it "Relay".
The Bluesky marketing is great at confusing people by giving them a false impression of decentralization: they let you run your own personal datastore with just your posts, and then the BGS will fetch your posts from that, no one else will connect to it, but that is just an unnecessary intermediary between you and the BGS, you could just publish directly to the BGS and I don't see why you would not do it.
