3 ms·

I can ping many popular destinations on the internet in <4ms. Why does a cross town video call have perceptible latency?

com2kid 6y ago
Oh its worse than that. Pick a popular video chat application, have you and someone else on the same wireless network both join the same chat, go into separate rooms to avoid crosstalk on the mics. Have one person talk, then realize there is a half a second lag. It doesn't help that many bluetooth headsets add up to 200ms of latency (!!) https://www.rtings.com/headphones/tests/connectivity/bluetooth https://www.rtings.com/headphones/tests/connectivity/bluetoo... We'd be better off using analog telephone lines and 900mhz wireless headsets from the 90s.
- closeparen 6y ago
  I wonder if there will ever be a chance to burn down this whole playback-oriented multimedia stack and build it back up for real time communication.
  - toast0 6y ago
    It's possible, but it's not likely to come from established players. And it's going to have to sacrifice visual quality or use a ton of bandwidth (or both). I don't know if video calling tends to multiplex audio and video, or send separate streams; multiplexing audio onto the video could help reduce audio delay though. Standard voip uses a packet rate of 50 pps for audio, which means at least 20 ms between sampling and sending. Reducing that would be nice.
    - com2kid 6y ago
      > It's possible, but it's not likely to come from established players. And it's going to have to sacrifice visual quality or use a ton of bandwidth (or both). Switching to Peer to Peer for 1:1 or small group chats would be nice. Low latency requirements being part of the BT spec would also be nice[1]. OS's fixing their drivers and entire end to end audio stack, that'd also be nice. Get that end to end latency locally to less than the e2e latency of the network packets! [1] Qualcomm has a low latency version of apt-x which is actually low latency, and Apple does fine with Airpods connected to Mac hardware.
      - closeparen 6y ago
        AirPods Pro connected to Mac hardware still run 144ms. That’s an eternity in live audio world, latency for a genuinely “pro” digital wireless microphone system is closer to 1.4ms.
        com2kid 6y ago
        Holy cow, I misremembered. That is... so bad. How? Did they let the same people who designed Bluetooth LE's ANCS[1] go hog wild with the desktop stack? Edit: Also I wish that site tested latency for phone calls, since that is often a different protocol. [1] ANCS is, or at least last time I looked at it, a protocol designed by people who had obviously NEVER worked in the embedded or low power space. No limit on message size and no headers indicating the size of the incoming payload are two giant tells.
gruez 6y ago
AFAIK the killer is (video) encoding latency, which in turn comes from a tradeoff between encoding efficiency (bandwidth use) and number of frames that need to be "buffered" inside the encoder. This is exhibited inside the videoconferencing software and the webcam itself (AFAIK webcams don't send an uncompressed stream to the computer).
rkangel 6y ago
Because packet based networks are the worst basis for real time communication. There is a good reason there's a whole separate international connection oriented network for making phone calls (it's a shame it's no good for video - unsurprisingly the internet wins on open platforms and innovation). There are two important differences with VC compared to that ping: You're not trying to exchange data with a CDN built for exactly that purpose, you're trying to exchange it with someone else's (probably crappy) internet connection. The analogy I used the other day is "your supermarket has better roads to it than your friend's house". You're probably pinging 32 or 64 bytes. Now try that again at a more realistic video data rate (not that the CDN endpoint will allow that). Keeping latency down is a lot easier on small packets than on full pipes.

com2kid 6y ago

Oh its worse than that. Pick a popular video chat application, have you and someone else on the same wireless network both join the same chat, go into separate rooms to avoid crosstalk on the mics. Have one person talk, then realize there is a half a second lag. It doesn't help that many bluetooth headsets add up to 200ms of latency (!!) https://www.rtings.com/headphones/tests/connectivity/bluetooth https://www.rtings.com/headphones/tests/connectivity/bluetoo... We'd be better off using analog telephone lines and 900mhz wireless headsets from the 90s.

closeparen 6y ago
I wonder if there will ever be a chance to burn down this whole playback-oriented multimedia stack and build it back up for real time communication.
- toast0 6y ago
  It's possible, but it's not likely to come from established players. And it's going to have to sacrifice visual quality or use a ton of bandwidth (or both). I don't know if video calling tends to multiplex audio and video, or send separate streams; multiplexing audio onto the video could help reduce audio delay though. Standard voip uses a packet rate of 50 pps for audio, which means at least 20 ms between sampling and sending. Reducing that would be nice.
  - com2kid 6y ago
    > It's possible, but it's not likely to come from established players. And it's going to have to sacrifice visual quality or use a ton of bandwidth (or both). Switching to Peer to Peer for 1:1 or small group chats would be nice. Low latency requirements being part of the BT spec would also be nice[1]. OS's fixing their drivers and entire end to end audio stack, that'd also be nice. Get that end to end latency locally to less than the e2e latency of the network packets! [1] Qualcomm has a low latency version of apt-x which is actually low latency, and Apple does fine with Airpods connected to Mac hardware.
    - closeparen 6y ago
      AirPods Pro connected to Mac hardware still run 144ms. That’s an eternity in live audio world, latency for a genuinely “pro” digital wireless microphone system is closer to 1.4ms.
      - com2kid 6y ago
        Holy cow, I misremembered. That is... so bad. How? Did they let the same people who designed Bluetooth LE's ANCS[1] go hog wild with the desktop stack? Edit: Also I wish that site tested latency for phone calls, since that is often a different protocol. [1] ANCS is, or at least last time I looked at it, a protocol designed by people who had obviously NEVER worked in the embedded or low power space. No limit on message size and no headers indicating the size of the incoming payload are two giant tells.

gruez 6y ago

AFAIK the killer is (video) encoding latency, which in turn comes from a tradeoff between encoding efficiency (bandwidth use) and number of frames that need to be "buffered" inside the encoder. This is exhibited inside the videoconferencing software and the webcam itself (AFAIK webcams don't send an uncompressed stream to the computer).

rkangel 6y ago

Because packet based networks are the worst basis for real time communication. There is a good reason there's a whole separate international connection oriented network for making phone calls (it's a shame it's no good for video - unsurprisingly the internet wins on open platforms and innovation). There are two important differences with VC compared to that ping: You're not trying to exchange data with a CDN built for exactly that purpose, you're trying to exchange it with someone else's (probably crappy) internet connection. The analogy I used the other day is "your supermarket has better roads to it than your friend's house". You're probably pinging 32 or 64 bytes. Now try that again at a more realistic video data rate (not that the CDN endpoint will allow that). Keeping latency down is a lot easier on small packets than on full pipes.