WritingScalewayScalewaypublished Mar 28, 2024seen 5d

How to optimize your cloud infrastructure for video streaming and encoding - Webinar report

Open original ↗

Captured source

source ↗

How to optimize your cloud infrastructure for video streaming and encoding - Webinar report Scale • James Martin • 28/03/24 • 4 min read

Today, the cloud is an essential part of video processing. Naturally, when you consider Netflix and YouTube alone accounted for over a quarter of global web traffic in 2022. Furthermore, that and related activities are set to grow: IT spending in the global media and entertainment sector rose by 34% between 2019 and 2024, according to Gartner .

What do these trends imply for companies in the sector today, particularly for video companies? We brought three of them together for a dedicated webinar late March 2024. Here are the highlights!

"I've been in the video sector for 18-20 years, notably on the VLC, FFmpeg and x264 projects," said Jean-Baptiste Kempf, president of VideoLAN and inventor of the pioneering VLC open source player, by way of introduction. "Today, all video found on the cloud and on websites is managed by open source projects, whether it’s on Facebook, YouTube or Twitch. This is one of the few areas of the cloud where there is virtually no non-open source competition, and all proprietary solutions are based on these solutions. But it's true that today you need significant resources for transcoding - so a lot of computing power - and adapted infrastructures that are capable of scaling according to your needs at certain times. So the importance of cloud providers is paramount."

Our three speakers then presented their solutions.

Emmanuel Fréard , Co-Founder and CTO of blastream , presented his cloud-based livestreaming service, which enables events to be produced and broadcast directly over the web. "It’s a sort of mini- OBS in the cloud, controlled directly from the web, similarly to what our American competitors, such as StreamYard, offer," explained Fréard. Blastream has the particularity of being white-label, which makes it suitable for uses such as webinars, but also live shows, as well as special activations like the broadcast of a remote-controlled car race, with Orange . To achieve this, blastream relies on MCU technologies, with ingestion and broadcasting notably via WebRTC. On the cloud side, blastream depends on the reliability and availability of Scaleway's CPU resources - "we have an average of 25 seconds of instance availability, which is very comfortable for our customers and end users", asserted Fréard - as well as the sovereign aspect, as video data must remain RGPD-compliant at all times.

Geoffrey Signorato , Lead Developer at Streamfizz , then presented this video streaming platform created in 2012, by parent company Webcastor. Streamfizz features a responsive, customizable player and alternative to GAFAM solutions, with a strong focus on data confidentiality. Indeed, Streamfizz's cloud configuration "ensures that videos are not sent to services outside France", explains Signorato; "this is why Streamfizz has developed its entire encoding system in-house" . This system consists of uploading video files to object storage, before passing them on to a pool of 3070 and RENDER-S GPUs in a private network for encoding and subtitling, before they go back to object storage. Streamfizz also relies on Scaleway's Kubernetes Kapsule to manage scale-up and scale-down of GPU Instances.

François Caron , CEO of EMPREINTE.COM , then explained how this "pioneer in online video" began in 1989, producing CD-ROMs for Wanadoo (France Télécom), before moving on to webTV, streaming, display and webinars. To do this, Empreinte relies on Scaleway Kapsule, Load Balancer, Object Storage, Container Registry and MySQL, enabling it to meet substantial load requirements, such as 70,000 simultaneous users for one customer , or 3,000 multilingual live broadcasts per year for another. Caron was particularly pleased that his sector had created 40,000 jobs to date - "we have talent in this country, and applications that are used worldwide" - while regretting that said players "don't talk to each other enough."

Video webinar speakers

Why sovereignty comes first with video

Sovereignty being the common ground between these three players, the discussion part of the webinar began on this topic. "It's a determining factor", insisted Caron, particularly when managing the video content of clients’ intranets, as is the case with Empreinte. "Some servers are in the US, which can reveal confidential information. CIOs don't want to look at this sort of thing... except when we talk to them about cybersecurity. Controlling cloud perimeters as tightly as possible is decisive for security."

It's also a key sales argument, according to Signorato. "It's a real selling point to be able to tell our customers that we know exactly where their video will go when they upload it . We know that it stays in France, who encodes it and where, where it's uploaded... Sovereignty is not something clients think about by themselves, but when we tell them about it, it tends to reassure them."

In addition to reassuring customers, sovereignty is "mandatory for some customers", according to Fréard. "Some CIOs impose it. And that's fine with us!” he said. “We prefer to work with people we can talk to, which isn't always the case for non-European companies. It's a sales pitch, but also a technical one, and one that has real added value for us as a startup."

Availability and flexibility: the keys to success

Fréard then returned to his need for guaranteed CPU Instances available within 25 seconds. "We don't want to have a whole pool of servers available at all times and only use them 10% of the time. We prefer to call on resources at a moment's notice" , he said. “This makes economic and environmental sense. And the faster these resources are available, the better it is for the end customer.”

This need for flexibility is covered for Streamfizz by Kubernetes autoscaling, "which allows us to define a minimum of replicates that will run on Instances", explained Signorato. "Whenever Kubernetes receives a CPU or RAM load, the autoscaler will automatically switch on Instances," he added. "This elasticity allows us both to manage peak loads, and to limit resources to the minimum" when Streamfizz doesn't need resources . “There's a whole configuration to be done in advance, but once properly set up, we have a certain robustness on production."

"We're trying to do better with less",…

Excerpt shown — open the source for the full document.