Call processing capacity — RTPengine in RELAY mode (CPS and concurrent sessions)

[Ricardo Deizeppe]

Hello everyone,

I’m currently running performance tests with RTPengine in RELAY mode, integrated with Kamailio.
I’d like to better understand the real-world processing capacity (in terms of CPS and concurrent active calls) that other operators or maintainers have experienced in production or lab environments.

Current test scenario:
• Mode: RELAY
• Number of dedicated CPU cores: 64
• num-threads / media-num-threads: 64
• SIP traffic generated using SIPp
• RTPengine running on bare metal (no virtualization)
• Kernel module xt_RTPENGINE loaded
• RTP port range: 20000–50000
• Plain UDP (no ICE/SRTP)
• Dedicated 10 Gbps interface for media traffic

My questions:
• What is the maximum practical CPS that RTPengine can sustain in this kind of pure relay scenario?
• Are there any recommended or observed limits for the number of simultaneous active calls per instance (before jitter, packet loss, or high latency start to appear)?
• Beyond thread and IRQ pinning tuning, are there any critical tuning parameters you recommend for reaching higher performance?

Thanks in advance for any benchmarks, tuning tips, or field experience you can share.