If this whole non-archival relay thing moves forward as is, it means pretty much any dev can run a relay. Requirements move down to ~8 cores, 16gb ram, ~2Tb ssd (can reduce this too) and as much bandwidth as you want to provide to your downstream users
Obv would scale with network, but a good start
Obv would scale with network, but a good start
Comments
But yeah the network was <2TB like 3 months ago including the CAR files of all repos.
I run two servers for like .. maybe 40 a month ?
Theres a reason we dont use cloud boxes
That's includes the electricity, cool ing, some backup, but not Internet bandwidth
“don’t fuck it up”
https://github.com/bluesky-social/indigo/blob/main/cmd/bigsky/README.md
the appView is pretty expensive to run for us, but it scales per user of our app, not necessarily with the size of the network.
The nice thing about making the relay cheap is that other app developers can set their own up easily
Each server can choose who to peer with.
Article IDs are used to stop forward propagation of dupes.
Kinda like how some activitypub services (e.g. Lemmy) let you federate post searches between instances - you can search the indexes of all your federated instances at once.
I still want to bridge those partitions somehow, and afaict that can happen either on the client or (as on fedi) by horizontal interop with a third "neutral" service
📌
(Yup, I just want Bsky to work like Twitter w/out looking under the hood).
It's been at least a decade or more of feeing increasingly isolated from participating in this stuff. The change feels good.
Do you have any notion for "how many per second" we're talking with those specs?
I had mistakingly thought it was more of a shared infra situation, like running your own load balancing instance for the benefit of the wider network