Broadcasting live events comes with a set of specific challenges that are simply not as severe when dealing with pre-recorded content. A group of experts came together to explain how to tune up your approach to get the best results, reports Andrew Williams.
Live event streaming raises its own, specific challenges, as well as some broader hurdles to surmount. Latency becomes a huge factor, particularly when streaming sport or auctions where only an experience close to real-time will suffice. The spikes in demand for events like football matches are a challenge for any broadcaster, making the management of server load critical.
The pitfalls of live event broadcasting have been a perennial IBC topic for some years, but the IBC2023 show saw a panel discussion between three companies with considerable expertise in solving these problems - Gcore, NPAW and VisualOn.
The talk – The Technical Challenges of Live Streaming - identified three key elements that tend to lead to greater success in the world of event live streaming - multi-CDN, dynamic CDN management and dynamic encoding. These three strategies can all help avoid technical headaches and improve the user experience of live streaming. Let’s take a closer look.
Live Event Streaming: Multi-CDN
In the simplest of streaming infrastructures, a single CDN, content delivery network, may be used. However, multi-CDN can make a streamer much more resilient as well as reducing latency.
If your audience is distributed across the world, you do not want them to have to stream from a single location when latency is a key factor.
“It’s a great disappointment when your neighbours are watching football on TV, and they see the greatest goal much faster than you who, like a modern man, watched the same stream on the internet,” said Anastazja Melnikova, Streaming Solution Architect at Gcore.
“The distance challenge is also a thing because the latency for broadcasting is very important. In theory every 96 kilometres of distance adds about one millisecond. So communicating with the server in another part of the world can add more than 2000 to 3000 milliseconds to the latency and we are talking only about connectivity latency here.”
By using a multi-CDN with servers located around the world, the overall latency can be dramatically reduced.
Melnikova said this location factor should be one of the prime considerations when looking for a CDN partner.
“Frankly speaking, every CDN provider has the same bunch of features,” she said.
“The main thing you should look for when finding the best CDN provider is a global presence and even the company location. [This is] because the huge CDN players were born in the USA but nowadays a lot of European customers, based on legal reasons and GDPR, want to have a native European provider.”
It’s not all down to matching the server location to that of your audience either. A multi-CDN array is ideal when dealing with the demand spikes that come with streaming of events, football matches being the key example that typically comes to mind.
“We have a global CDN, and usually sports viewers are located globally. So, because of that, different CDN servers will answer the demand… and we also have clusters of CDN in each location. If one server is down, the other server will answer and there is balancing between them. So not all the viewers will come to one server,” explained Melnikova.
Live Event Streaming: Dynamic CDN
Luis López Chousa, Head CDN Balancer at NPAW was one of the original advocates for a multi-CDN approach.
“12 years ago, at the very beginning [of NPAW], were trying to convince customers to have a multi CDN approach… There was almost no customer having multi CDN 10 years ago, Mainly the CDNs were giving a much better price to customers when they got all the volume on a single CDN.”
Working with multi-CDN systems for more than a decade has led Chousa to think more deeply about how they can and should function.
He explained the difference between a standard approach, where a CDN is selected at the beginning of the stream, and what he calls active switching. This is when the viewer may be moved to a different CDN’s server mid-stream.
“When something is about to go south, there’s a change in CDN that avoids buffering,” said Chousa. “It’s seamless for the end user.”
NPAW’s testing found active switching can reduce buffer ratio by 50%, start-up errors by 91% and in-stream errors by 42%. This was through the use of just two CDN instances, switching between them to avoid playback interruptions.
To make sure this active switching works in a live setting without added latency, it simply has to be ensured that each CDN uses the same encoding, in order to make the packets of data the CDNs store identical.
Live Event Streaming: Dynamic encoding
That exact process of encoding that data can have a huge impact on a streamer’s ability to handle load, and the cost involved in live streaming.
VisualOn is a specialist in the area, employing the latest techniques to dramatically reduce the data demand of live-streamed video without impacting visual quality.
It involves the hot tech topic of the moment, artificial intelligence, as Christophe Coquerel, Director of Application Engineering at VisualOn, explained. “You need to detect changes at the frame level,” said Coquerel.
“We take sample frames every X frames with a machine learning and AI model that will detect, ok we have a change, and work out the best settings for the encoder and apply them on-the-fly.”
“Not each frame needs the same quantity of bits for the same quality level.”
There’s a fairly intense analysis of the content itself, looking at what’s actually seen in camera, to determine what encoder settings are required to hit the desired quality level, per scene.
This kind of more dynamic encoding method seems entirely sensible for standard OTT content but bringing that level of smart engineering into live streaming without introducing additional latency is the key innovation here. There’s “no added latency for live” streaming, according to VisualOn.
Coquerel said the company’s technology can achieve “average savings in the 50% range” for data use, in a classic example of “minimum impact for immediate results.”
This AI-informed encoding technique can of course be used to reduce load on a stream’s CDN network, to let it better deal with the spikes of a popular live streamed event. However, it can just as easily be used to dramatically improve image quality, while keeping the same data rate.
Coquerel also said VisionOn’s technology does not require a major re-think of system architecture. “We don’t, at all, impact the delivery chain,” he said.
Smart encoding, multi-CDN and smart switching see the building blocks of streaming infrastructure levelled up, to let them better cope with the specific challenges of live streaming, and it doesn’t take too much deep thought to see how they could work together in harmony.
Colin Dixon, Founder and Chief Analyst at nScreenMedia chaired the panel discussion on the technical challenges of live streaming at IBC2023. The panel featured Anastazja Melnikova of Gcore, Luis López Chousa of NPAW and Christophe Coquerel from VisualOn.