3 ms·

Maybe Twitch's management effectively prefers 3s of latency at 20x the bandwidth cost of 7 seconds of latency, but I have to admit that it would surprise me. T

Maybe Twitch's management effectively prefers 3s of latency at 20x the bandwidth cost of 7 seconds of latency, but I have to admit that it would surprise me. Twitch is an interactive medium, not merely a video stream delivery platform. Streamers chat with their viewers and play games with them. They run merchandise giveaways based on guessing games. Low latency is critical for these applications. If a stream viewer sees a ton of people guessing random words or numbers in chat before they even see the streamer mention what everyone’s guessing at, they’re going to have a bad experience, leading to disengagement. Twitch goes out of their way to lower the latency. They even offer options in the client to switch to a lower latency mode, at the expense of stability.

littlestymaar 3y ago
I was a bit confused when reading your comment at first because I was a big Twitch user at a time when they had around 10-15s latency for everyone without feeling that this latency was a problem. But then I eventually understood that you're talking about latency between viewers and then indeed it would be a very frustrating experience. Fortunately it's not what I'm talking about: I'm talking about having some latency (or rather “buffer size”) for all users, with synchronized playback between viewers. And again we're not talking about a prohibitive buffer size but rather a 5-15s buffer size, which is pretty common for live streaming services.
- chongli 3y ago
  How does that work? Let’s say a stream is running with a 5s latency with the streamer and a bunch of viewers, all located in the US on fast connections. Now someone from rural Australia connects to the stream on a slow connection. Do you pause the stream for everyone in order to build a buffer for that new user and allow them to catch up and synchronize? What does the latency look like after that? 15-20s for everyone? That sounds like a terrible experience! When users find out what’s going on they’re going to demand the ability to block these slow/distant users from the stream so that everyone else doesn’t have their performance degraded. It also doesn’t solve all of the interactivity issues. If a streamer is directly chatting with their viewers, a 15s latency is huge! It effectively makes normal back-and-forth conversations impossible and therefore leads to disengagement. Why is this such a problem? Look at the relationship between engagement and monetization. Highly engaged viewers give way more money to streamers than do passive viewers. Highly engaged viewers are simply way more valuable as an audience, so their experience is given top priority. This is why best-effort low latency without synchronization makes sense. Lowest-common-denominator synchronized video sounds much more difficult to achieve if these “power viewers” are your priority audience.
  - littlestymaar 3y ago
    > How does that work? Let’s say a stream is running with a 5s latency with the streamer and a bunch of viewers, all located in the US on fast connections. Now someone from rural Australia connects to the stream on a slow connection. It works exactly as for non-p2p streaming, as the playback synchronization is independent from the p2p part and entirely managed by the video player through its buffer management system. Basically you get the “latest block” from the stream's manifest (that always comes from the CDN) and then the player is playing a few blocks behind and keeping its buffer full. > What does the latency look like after that? 15-20s for everyone? Yes, but that's already the case for most live streaming website you can find unless they specifically aim for low-latency (which comes with a QoS drawback since smaller buffer means buffering is more likely for the end user if the connection isn't flawless, so going to low-latency is a trade-off) and it was the case on Twitch itself before 2019. > It also doesn’t solve all of the interactivity issues. If a streamer is directly chatting with their viewers, a 15s latency is huge! It effectively makes normal back-and-forth conversations impossible and therefore leads to disengagement. It's less convenient, but at the same time Twitch has worked like that for almost a decade so it's not a complete showstopper. And again, 15s isn't the minimum you can reach, it's going below 5s that's not necessarily feasible (also, when I left my former employer, 5s was the limit but people were working on that so maybe they managed to reach 3s as they were trying to). > Highly engaged viewers are simply way more valuable as an audience, so their experience is given top priority. This is why best-effort low latency without synchronization makes sense. Maybe, but how many times more valuable? Has the per user value increase by a factor 20 since 2019? If no, maybe having reduced the bandwidth cost would have made more sense that trying to improve engagement.
    - chongli 3y ago
      It works exactly as for non-p2p streaming, as the playback synchronization is independent from the p2p part and entirely managed by the video player through its buffer management system. Basically you get the “latest block” from the stream's manifest (that always comes from the CDN) and then the player is playing a few blocks behind and keeping its buffer full. Those blocks still need to be sent to the new user's entry point to the CDN. I doubt you're going to save any money by sending every single video stream to every single node in the CDN. You're going to send the video only to nodes that have connected viewers of that particular stream. Then you have the issue of pausing everyone else’s feed until the new user catches up. Yes, but that's already the case for most live streaming website you can find unless they specifically aim for low-latency (which comes with a QoS drawback since smaller buffer means buffering is more likely for the end user if the connection isn't flawless, so going to low-latency is a trade-off) and it was the case on Twitch itself before 2019. I think Twitch has moved on from that. Users' expectations have gone way up. They want more interaction with streamers, not less. They aren't going to tolerate a backslide to higher latencies if the only purpose is to save Twitch money. Competition from other platforms, especially YouTube, is fierce. High latency (>5s) leads to this shallow form of engagement where viewers send one-way comments to the streamer and then wait (and hope) they hear back. Low latency promotes live conversations. This is especially critical on small streams where the number of people in the chat doesn't overwhelm the streamer. And as I said before, smaller streams are less likely to have streamers who lean into heavy third-party monetization strategies. These streamers often develop closer relationships with their viewers and have a higher percentage of subscribed viewers. Maybe, but how many times more valuable? Has the per user value increase by a factor 20 since 2019? If no, maybe having reduced the bandwidth cost would have made more sense that trying to improve engagement. I think they are trying to reduce bandwidth costs, but not uniformly across users. My theory is that they are trying to discourage casual viewers who don't subscribe. Getting rid of these "freeloader users" would naturally cause per-user value to go up!
- jorams 3y ago
  > But then I eventually understood that you're talking about latency between viewers The latency between streamer and viewer on Twitch (in low-latency mode, which is the default) is normally less than 5 seconds and often between 1 and 2 seconds. The player actually slightly speeds up the video if it exceeds 5 seconds. This is essential for interactivity, because the streamer can ask a question and get a response without constant awkward pauses.
  - littlestymaar 3y ago
    It's only been the case since 2019 so I dispute the characterization as “essential for interactivity” since Twitch has run for almost a decade without it. Maybe it improves things but we're back to my previous comment: > Maybe Twitch's management effectively prefers 3s of latency at 20x the bandwidth cost of 7 seconds of latency, but I have to admit that it would surprise me.

littlestymaar 3y ago

I was a bit confused when reading your comment at first because I was a big Twitch user at a time when they had around 10-15s latency for everyone without feeling that this latency was a problem. But then I eventually understood that you're talking about latency between viewers and then indeed it would be a very frustrating experience. Fortunately it's not what I'm talking about: I'm talking about having some latency (or rather “buffer size”) for all users, with synchronized playback between viewers. And again we're not talking about a prohibitive buffer size but rather a 5-15s buffer size, which is pretty common for live streaming services.

chongli 3y ago
How does that work? Let’s say a stream is running with a 5s latency with the streamer and a bunch of viewers, all located in the US on fast connections. Now someone from rural Australia connects to the stream on a slow connection. Do you pause the stream for everyone in order to build a buffer for that new user and allow them to catch up and synchronize? What does the latency look like after that? 15-20s for everyone? That sounds like a terrible experience! When users find out what’s going on they’re going to demand the ability to block these slow/distant users from the stream so that everyone else doesn’t have their performance degraded. It also doesn’t solve all of the interactivity issues. If a streamer is directly chatting with their viewers, a 15s latency is huge! It effectively makes normal back-and-forth conversations impossible and therefore leads to disengagement. Why is this such a problem? Look at the relationship between engagement and monetization. Highly engaged viewers give way more money to streamers than do passive viewers. Highly engaged viewers are simply way more valuable as an audience, so their experience is given top priority. This is why best-effort low latency without synchronization makes sense. Lowest-common-denominator synchronized video sounds much more difficult to achieve if these “power viewers” are your priority audience.
- littlestymaar 3y ago
  > How does that work? Let’s say a stream is running with a 5s latency with the streamer and a bunch of viewers, all located in the US on fast connections. Now someone from rural Australia connects to the stream on a slow connection. It works exactly as for non-p2p streaming, as the playback synchronization is independent from the p2p part and entirely managed by the video player through its buffer management system. Basically you get the “latest block” from the stream's manifest (that always comes from the CDN) and then the player is playing a few blocks behind and keeping its buffer full. > What does the latency look like after that? 15-20s for everyone? Yes, but that's already the case for most live streaming website you can find unless they specifically aim for low-latency (which comes with a QoS drawback since smaller buffer means buffering is more likely for the end user if the connection isn't flawless, so going to low-latency is a trade-off) and it was the case on Twitch itself before 2019. > It also doesn’t solve all of the interactivity issues. If a streamer is directly chatting with their viewers, a 15s latency is huge! It effectively makes normal back-and-forth conversations impossible and therefore leads to disengagement. It's less convenient, but at the same time Twitch has worked like that for almost a decade so it's not a complete showstopper. And again, 15s isn't the minimum you can reach, it's going below 5s that's not necessarily feasible (also, when I left my former employer, 5s was the limit but people were working on that so maybe they managed to reach 3s as they were trying to). > Highly engaged viewers are simply way more valuable as an audience, so their experience is given top priority. This is why best-effort low latency without synchronization makes sense. Maybe, but how many times more valuable? Has the per user value increase by a factor 20 since 2019? If no, maybe having reduced the bandwidth cost would have made more sense that trying to improve engagement.
  - chongli 3y ago
    It works exactly as for non-p2p streaming, as the playback synchronization is independent from the p2p part and entirely managed by the video player through its buffer management system. Basically you get the “latest block” from the stream's manifest (that always comes from the CDN) and then the player is playing a few blocks behind and keeping its buffer full. Those blocks still need to be sent to the new user's entry point to the CDN. I doubt you're going to save any money by sending every single video stream to every single node in the CDN. You're going to send the video only to nodes that have connected viewers of that particular stream. Then you have the issue of pausing everyone else’s feed until the new user catches up. Yes, but that's already the case for most live streaming website you can find unless they specifically aim for low-latency (which comes with a QoS drawback since smaller buffer means buffering is more likely for the end user if the connection isn't flawless, so going to low-latency is a trade-off) and it was the case on Twitch itself before 2019. I think Twitch has moved on from that. Users' expectations have gone way up. They want more interaction with streamers, not less. They aren't going to tolerate a backslide to higher latencies if the only purpose is to save Twitch money. Competition from other platforms, especially YouTube, is fierce. High latency (>5s) leads to this shallow form of engagement where viewers send one-way comments to the streamer and then wait (and hope) they hear back. Low latency promotes live conversations. This is especially critical on small streams where the number of people in the chat doesn't overwhelm the streamer. And as I said before, smaller streams are less likely to have streamers who lean into heavy third-party monetization strategies. These streamers often develop closer relationships with their viewers and have a higher percentage of subscribed viewers. Maybe, but how many times more valuable? Has the per user value increase by a factor 20 since 2019? If no, maybe having reduced the bandwidth cost would have made more sense that trying to improve engagement. I think they are trying to reduce bandwidth costs, but not uniformly across users. My theory is that they are trying to discourage casual viewers who don't subscribe. Getting rid of these "freeloader users" would naturally cause per-user value to go up!
jorams 3y ago
> But then I eventually understood that you're talking about latency between viewers The latency between streamer and viewer on Twitch (in low-latency mode, which is the default) is normally less than 5 seconds and often between 1 and 2 seconds. The player actually slightly speeds up the video if it exceeds 5 seconds. This is essential for interactivity, because the streamer can ask a question and get a response without constant awkward pauses.
- littlestymaar 3y ago
  It's only been the case since 2019 so I dispute the characterization as “essential for interactivity” since Twitch has run for almost a decade without it. Maybe it improves things but we're back to my previous comment: > Maybe Twitch's management effectively prefers 3s of latency at 20x the bandwidth cost of 7 seconds of latency, but I have to admit that it would surprise me.