What is the average latency in Twilio Video Rooms?

dtoomey admin
edited June 7 in Video & WebRTC

Short answer:
Latency does not depend on Twilio but on the device and network access type of the end user. Hence, in general, it does not make sense to provide “average latencies” or “latency metrics” because the latency of a user connecting through an Ethernet link will be completely different than the latency when connecting from a 3G network or a satellite link.

Long answer:
In a Group Room the overall latency is the sum of three factors:
1. The network latency (propagation delay)
2. The media server latency (time the information is "inside the media server")
3. The jitter buffer latency (time the information is buffered at the receiver side before rendering.

In a P2P Room or WebRTC Go Room only factors (1) and (3) count.

Regarding (1), it depends on the type of access of the clients and on their geographical location. For example:

  • When access happens through 3G/4G networks latencies can grow significantly. We have seen "extra latencies" of up to 500ms added by these types of networks.
  • When clients are on opposite sides of the world, latency will increase. Latency from Australia to US may be quite variable (dependent on Internet status) and oscillate between 300 and 900ms easily.
  • All in all, for calls inside US, (1) may probably oscillate between 50ms (when network access is reasonable) and 500ms, when connecting through 3G/4G

Regarding (2), this one is typically low, around 10ms. It's the only one Twilio can control and it's typically very small.

Regarding (3)
The jitter buffer is a buffer the receiver side creates for compensating packet loss (waiting for retransmissions), packet disorder and packet jitter. We use a dynamic jitter buffer that adapts to the network conditions. When the network behaves correctly, it's typically quite small (<50ms) but when the network has lots of packet loss it may grow significantly.

All in all, what we can say is:

  • If the network is OK, our typical latency for calls inside us is under 250ms, which in general provides a very satisfactory end-user experience.
  • If client uses high latency links (e.g. 3G, satellite) the latency will grow adding such extra latency and Twilio cannot avoid it.
  • If the network has a lot of packet loss or a lot of jitter, the latency can also grow significantly an it's not rare to see latencies >500ms and more.
  • Latency is just another network property, like bandwidth. We manage latency and bandwidth similarly and we monitor both in our Network Quality Score. If latency grows degrading the QoE then the quality score will drop, even if BW is right.
Sign In or Register to comment.