What are the backend protocols for federation? I wonder if it’s true server load (physical lack of compute) or something like network bottlenecks (which we might be able to see by using things like NIC utilization and error rates).
Network bottlenecks may be addressable at the protocol layer with improvements to how federation works.
I’m pretty damn sure its physical lack of compute, because lemmy.ml already massively upgraded their server two weeks ago to account for this, and it’s already overloaded again.
Also, this was an issue early on in Mastodon as well. The more federated servers there are, the more each server has to communicate with every other server which apparently eats up RAM and CPU cycles. So each time there’s an update, it has to sent it to X number of servers, and retrieve data from X number of servers, and as X climbs, the more network activity and CPU cycles to handle it there is. A lot of requests in waiting can be stored away in RAM, but even RAM has limits.
Finally, the newest server version was supposed to help ease this kind of stuff by moving from using WebSockets to HTTP. Pretty sure the federation protocols are all just HTTP, but I could be wrong.
What are the backend protocols for federation? I wonder if it’s true server load (physical lack of compute) or something like network bottlenecks (which we might be able to see by using things like NIC utilization and error rates).
Network bottlenecks may be addressable at the protocol layer with improvements to how federation works.
I’m pretty damn sure its physical lack of compute, because lemmy.ml already massively upgraded their server two weeks ago to account for this, and it’s already overloaded again.
Also, this was an issue early on in Mastodon as well. The more federated servers there are, the more each server has to communicate with every other server which apparently eats up RAM and CPU cycles. So each time there’s an update, it has to sent it to X number of servers, and retrieve data from X number of servers, and as X climbs, the more network activity and CPU cycles to handle it there is. A lot of requests in waiting can be stored away in RAM, but even RAM has limits.
Finally, the newest server version was supposed to help ease this kind of stuff by moving from using WebSockets to HTTP. Pretty sure the federation protocols are all just HTTP, but I could be wrong.