hey folks for self hosted deployement how are you guys doing the scaling of concurrency and what sort of cmpute instance is needed for 100, 1000, 100k concurrency? can yous help me on this and also what is the recommended vcpu and memory size for one concurrent connections?

In my testing, a 4cpu instance can do 250 concurrent rooms, its mainly cpu bound. This is for streaming video, only audio should be more.

not the lk infra, but voice agents worker part

ok I don't have the numbers for the agents, but depends quite a bit on what they do, usually vad/turn detection takes the most out of the cpu

@flat-photographer-63184 Hi, I also have a 4 CPUs server but my agent is picking only 10-12 calls per agent How you are scaling 250 concurrent rooms without latency and VAD issues on 4 CPUs server? how many agents do you run? also are you using selfhost livekit or livekit cloud for this ?

@cuddly-bird-77915 the 250 rooms is for self-hosted livekit server, not for the agent - its for video streaming solution, doesn't use agents. 10-12 sounds about right with VAD, what CPU usage do you see at peak?