from my limited past experience websockets have been hard to scale just because connections are persistent. So having 10k users requires hardware resources for 10k users.
so I assumed that typically paid relays would want to limit the amount of open connections and thus limit this to only paying users.
if I am wrong or if this not a problem in foreseeable future then I don't need to further think about it :D
Weāve been scaling Eden horizontally to handle more concurrent clients. But a lot can be done by making Nostream more performant using LMDB. Weāll get there in due time.
I would say regardless of exactly where the bottleneck is for the sake of his conversation you absolutely will see some paid relays use client auth when it is widely available. Otherwise your large āreadā costs will always be heavily subsidized by a small pool of writers. Long term those economics wonāt work for most paid relay operators.
This is a problem that will have to be dealt with but I donāt know that itās unreasonable for the short term solution to just be you may not (necessarily) receive 9735s on those relays.
We could probably kick that can further by making subscriptions more efficient. Most clients just get the same stuff back over and over.
how many concurrent clients do you currently have roughly there? on what resource usage?
