Hi there,
I'm running TURN server in a production environment on AWS. Individual server is currently taking 800-1000 sessions at peak time on a daily basis for media traffic. Each session on average is around 1Mbps.
I have observed udp packet loss counted in /proc/net/udp as UdpRcvbufError, it seems random and sometimes spikes to 1000.
I'm wondering if any known kernel tweak can optimize the server? I have been trying AWS c5n.18xlarge, as well as c5n.2xlarge. Not much difference.