179
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
this post was submitted on 15 Jun 2023
179 points (93.7% liked)
Programming
17314 readers
157 users here now
Welcome to the main community in programming.dev! Feel free to post anything relating to programming here!
Cross posting is strongly encouraged in the instance. If you feel your post or another person's post makes sense in another community cross post into it.
Hope you enjoy the instance!
Rules
Rules
- Follow the programming.dev instance rules
- Keep content related to programming in some way
- If you're posting long videos try to add in some form of tldr for those who don't want to watch videos
Wormhole
Follow the wormhole through a path of communities !webdev@programming.dev
founded 1 year ago
MODERATORS
Pulling this out of my ass, but I think the problem might be in Lemmy using websockets.
I feel like supporting 1500 simultaneous users making a request every 10-20 seconds is easier than keeping 1500 websockets alive.
Irregardless, Lemmy does feel very snappy compared to other websites I've had the displeasure of using. Main problem is low robustness in the RPC layer.
I maintain and host ntfy.sh, an open source push notification service. I have a constant 9-12k WebSocket and HTTP stream connections going, and I host it on a two core machine with an average load average of less than 1. So I can happily tell you that it's not WebSockets. Hehe.
My money would be on the federation. Having to blast/copy every single comment to every single connected instance seems like a lot.
As far as I know, every server connects to every other server. Allowing for proxying messages through servers would significantly help.
I agree.
Random ideas:
The Kademlia protocol (a DHT) has a thing that associates ownership of data to the 20 closest nodes in a P2P network. If an approach like this were used, the load would be spread across those 20 nodes. I implemented that like 15 years ago or so. It was a ton of fun.
Another, simpler approach is what you suggested, simple caching of and relaying through other nodes, though that does not answer the topology of the network. How would an instance decide where to get it's data from (a star, a tree, at random, ...)? How would it be authenticated (easy to solve)? Lots of fun problems to solve. Not fun problems though if you have a pile of other problems too though...
I thought of it like this:
If an instance misbehaves by not relaying messages despite claiming to be doing so - unfriend it.
Each instance publishes a public key that you can use to verify relayed messages.
I probably should get on to helping out developing Lemmy - it feels like there's RFC's to be written and interesting problems to be solved. Much more interesting than what I'm doing at work.