I’m running a test with Talk, which looks soooo great but, although chat is OK, I couldn’t launch a one-to-one audio/video communication. It seems the server lags and I’m suspecting that Talk puts a lot of processing power requirement on my poor virtual server (however, running a 2.4Ghz 4-core Xeon virtual processor).
Therefore, I’m wondering if there’s any abacus that would tell me: 2 connections = min RAM / min BPS, 10 connections = min RAM, min BPS, etc.
How many users use Talk? Do you run a TURN/STUN server? It helps a lot if your participants are behind restrictive firewalls/NAT networks.
Your hardware seems to be quite powerful for a lot of users. Can you run a tool to monitor your server performance? Netdata or sometimes a simple htop/nettop can show you the resource usage on your server.
Aside from that, I measured 2Mbit for each participant if a TURN server is used for Talk.
If you have users behind NAT or restrictive firewalls, then a TURN/STUN server is required for reliable operation. You can use Coturn which is well supported and you can find tutorials in other threads.
Your chat messages showing up with a high latency sounds like a strange problem. Maybe you should report it as a bug if it persists.
Thanks a lot Alfred! I’ll dig a bit about coturn… and will probably come back with new questions!
As for the chat latency, we’re currently migrating to our own dedicated server with plenty of cores/RAM/HDD, so I’ll wait to check the latency on this target infra…