Apparently (Gemini says https://dl.acm.org/doi/epdf/10.1145/3768994 was used to conclude), the top servers get 1 event per second, so I have to retract my earlier point about servers being able to be decommissioned.
Perhaps, if I may suggest-- how about this:
- A hardcoded list of, say, 5 relays with every installation.
- An introducer also chooses some b-tier nostr relay . In absence of an introducer, two devices randomly choose some b-tier nostr relay. Probably a vast majority of Syncthing graphs are disjoint anyway. This b-tier will be the highest priority relay for his disjoint graph. This way, Syncthing will be playing ideal digital citizen to not overload a single server. In case this relay is not working, the other 5 mainstream relays are used.
Should be noted though that nostr is up and coming and only increasing in popularity. Also, the infrastructure is currently underutilized clearly, 1 request per second when Jack Dorsey has given them $10 million (all the more reason for one single person to not bear the server cost). Also, an unsuspecting server that doesn’t want will have 30078 events dropped silently.