What Is the Best Chunk Size for Low-Latency Live Streaming?
The best chunk size for low-latency streaming is dependent on a number of factors based on different use cases, and there is often a need for compromise and tradeoffs. Nadine Krefetz, Consultant, Reality Software, Contributing Editor, Streaming Media, asks three industry experts what their chunk size preferences are for the particular needs of their platforms.
“We have somewhere like four to six seconds right now,” says Pankaj Chaudhari, Architect - Video Delivery, Disney Streaming. “There are many different tradeoffs to consider here. And though if the chunk size of the segment size is too low, obviously you are trying to get to a lower hand waving latency, glass-to-glass latency. The tradeoff is that now you have many connections, many requests are being sent to the protocol, the workflow becomes a little bit chatty, there are many numbers of requests per second that are coming in. But then at the same time, you are inserting a lot of keyframes in the segments. Every segment starts with a keyframe. And so the smaller the segment size, the more keyframes there will be over the course of the playback. And that may not be efficient from a cost point of view because there are more numbers of bits being delivered.”
Chaudhari also notes that Disney does not really have a use case for low latency. “Which is where WebRTC is not in use for us,” he says. “[So] four to six [seconds] kind of finds the nice, sweet spot wherein we have a balance of a good number of bytes being added to the payload, to the segments from a coding point of view, but also achieve reasonable latency.”
Krefetz says to Jonas Ribeiro, Digital Products, Platform and AdTech Manager, Globo, “You've got a lot of different things that you're supporting. Why don't you give us a little bit of background on that?”
“The live streaming size is six or seven seconds,” Ribeiro says. “We have a battle here because I do the TV broadcast and also do the live streaming. I believe that it is easier to delay the broadcast to be the same time as the live streaming, but this is not possible. For soccer games, we have a product that we call Premiere.” He says that this is a more expensive product due to higher technical requirements. It delivers in two-second chunk sizes to reduce latency. “It's just to give a better experience to these subscribers and it works well,” he says. “But it's not the same when you're seeing the broadcast and the live streaming, and I believe that is going to be a quite challenge, because it's a tradeoff.”
Imran Maskatia, VP of Product Development, Evoca TV, talks about the chunk size requirements for Evoca. “We use two-second segment sizes,” he says. “The reason we do so is for quicker channel change because as a TV service, we find a lot of people flipping channels, and the ability to start more quickly when you do that is kind of important for our customers.” He also mentions that he recently read a study out in South Korea. “They did a whole bunch of experimentation and looked empirically at everything from like a quarter second to five, ten seconds. And supposedly…they said that the optimal segment size tradeoff both for bandwidth, for fast channel change, etc., was somewhere around 0.9 seconds.”
“And that’s in the ATSC 3.0 environment?” Krefetz asks.
“That’s in the ATSC 3.0 context, right, for channel change,” Maskatia says. “So obviously there's a particular scope and use case in mind there, but it's a tradeoff.”
Learn more about chunk size and low-latency streaming at Streaming Media West 2022.
Watch full-session videos from Streaming Media Connect 2022.