Voices of Video
Explore the inner workings of video technology with Voices of Video: Inside the Tech. This podcast gathers industry experts and innovators to examine every facet of video technology, from decoding and encoding processes to the latest advancements in hardware versus software processing and codecs. Alongside these technical insights, we dive into practical techniques, emerging trends, and industry-shaping facts that define the future of video.
Ideal for engineers, developers, and tech enthusiasts, each episode offers hands-on advice and the in-depth knowledge you need to excel in today’s fast-evolving video landscape. Join us to master the tools, technologies, and trends driving the future of digital video.
Voices of Video
From AWS to On-Prem: Jet-Stream's Efficient Video Processing Revolution
The streaming industry is undergoing a significant transformation, particularly in Europe, where data sovereignty has become a critical concern for media companies. In this enlightening conversation, Stef van der Ziel, CEO of Jet-Stream, shares how his company has adapted to these changing demands by developing both sovereign cloud services and innovative on-premises solutions.
Stef takes us through the evolution of Jet-Stream from its early days in 1994 to its current position as a provider of European sovereign cloud services and on-premises video processing systems. What's particularly fascinating is how European regulations around data privacy have created new opportunities - requiring not just European hosting, but complete European ownership and operation with no connections to entities outside the region.
The discussion reveals a surprising industry trend: after years of migration to public cloud platforms, many organizations are returning to on-premises infrastructure due to unsustainable cloud costs. "The promise of the cloud was that your internal IT operation cost would go down by half," Stef notes, "and that's not true." This economic reality has driven Jet-Stream to develop MaelStrom, an innovative "micro cloud" solution that brings cloud-like scalability and resilience to on-premises deployments.
The technical implementation details are impressive - using NETINT VPU acceleration cards in ARM-based servers, a single 500-watt machine can process up to 30 full HD channels simultaneously with complete ABR ladders. This represents a 90% reduction in energy consumption compared to traditional approaches. All while maintaining high availability through containerized processing that automatically redistributes workloads if hardware fails.
Whether you're managing streaming infrastructure, concerned about data sovereignty, or looking to optimize your video processing costs, this episode offers valuable insights into how the industry is evolving to meet these challenges. Jet-Stream's approach demonstrates that with the right architecture, organizations can achieve the best of both worlds - cloud-like flexibility with the economics and control of on-premises infrastructure.
Subscribe to Voices of Video for more in-depth conversations about the technologies and trends shaping the future of video processing and delivery.
Stay tuned for more in-depth insights on video technology, trends, and practical applications. Subscribe to Voices of Video: Inside the Tech for exclusive, hands-on knowledge from the experts. For more resources, visit Voices of Video.
Voices of Video. Voices of Video. The Voices of.
Speaker 2:Video Voices of Video.
Speaker 1:Well, welcome back to another special edition of Voices of Video. So our guest today. I have talked with Steph. I think you've been on the show at least two times, maybe three times, right? Yeah, we always have a good time when you come on and you tell us the latest with Jetstream. So welcome to Voices of Video. It's great to have you back.
Speaker 2:Thanks to be here, Mark. Thank you very much.
Speaker 1:Yeah, you bet, you bet. So you know, when we were preparing for this, a lot has been happening, I know, with Jetstream over the last you know, year or so. Why don't you give us a quick update on, you know, the state of the business, state of the company, what you're focused on, and then let's talk about the cloud, but first, you know, update us on Jetstream.
Speaker 2:Right, okay, with Jetstream, we've been transforming from pure innovation to now doing business and scaling up into the market and, of course, the streaming market has been quite difficult for a lot of vendors, especially in the CDN space. 2024 was a tough year. It was a tough year for everyone and we believe that through innovation, you can keep moving on right and find new businesses and new customer bases and audiences. So one of the things we did was full focus on data protection and privacy. We are European and Sovereign Cloud is a major thing in Europe right now because of geopolitical issues. We won an award last year because of geopolitical issues. We won an award last year it's called the Dutch Privacy Award because we built a sovereign cloud in European ownership, european hosted, european operations.
Speaker 2:We basically phased out all the non-native services from third parties and moved everything in-house. So we built our own physical cloud with transcoding, packaging, player analytics Everything is in-house technology to guarantee customers that their data is safe in our secure cloud. Yeah, that's right. It's been hard to sell it in the beginning, but now suddenly we see a pickup because of politics. People are worried about their data, control of their data and who's accessing. Basically, who's watching what you're watching right, so I see a strong trend in there. And well, of course, transcoding has been one of those things that we wanted to move in-house. We outsourced this to AWS, but we've been running this for some years in-house right now, and, of course, one of the innovations was the hardware acceleration by NetEnt.
Speaker 1:Yeah, yeah, that's great. I mean we're really happy to be contributing, you know, to your cloud service, your cloud platform. So maybe what you should do is just, you know, kind of a one minute overview for those listeners and those viewers, those watching us on video, who actually aren't familiar with your product suite. You know, even with the history of the company. So just give a quick overview of you know who Jetstream is, how you started and what the core offer is. Okay.
Speaker 2:We started in 1994, streaming live on the web you are an OG, so still alive, yeah. And those events got bigger. So basically we crashed the Internet here in the Netherlands by overloading it and we had to figure out ways to scale it up. And that was a great mission. So I started the company with the mission to professionalize the web for streaming. Well, you know, 30 years later, here we are. I mean, who isn't watching streams on the internet? Right, that's right On mainstream.
Speaker 2:What we do is we offer two products. One is a European cloud service purely for streaming. So it includes multi-CDN services with integrations with a lot of CDNs, and it does origin services like transcoding, packaging, origin hosting, porting player and analytics coding, packaging, origin hosting, porting player and analytics. And on top of that we have an additional layer with APIs and GUIs to automate those workflows. So it goes beyond what AWS and Azure, for instance, are doing with streaming. They basically offer you a toolkit.
Speaker 2:We offer pre-integrations with full control. So it's a standard platform on which you can create custom workloads much faster and way cheaper than you used to do. And the second product is an on-prem solution. So there is another trend in which people say forget about the cloud. We want to go back into owning or at least operating on-premises our own infrastructure, like hospitals or defense or government, and they're so critical about either the solution or so critical about the data that they don't want to house this in any third-party cloud. So we built a small version of our cloud stack around transcoding packaging, origin hosting called Maelstrom.
Speaker 1:Yeah, it's a really interesting know. There's plenty of platforms and service providers out there who are bringing yet another SaaS type solution right Cloud, and there's a place for those. We're not in any means saying there isn't, but it's a tough space to be in. But you are correct, it's a tough space to be in, but you are correct. In Europe, maybe it's GDPR and it's the privacy that possibly is initially, I think, driving. You know your traffic cannot leave this region, you know, and it has to be processed only within the region that the consumers are in, the viewers are in, and so that means then you know, suddenly, aws doesn't necessarily work so well in that type of environment right, it's a great platform.
Speaker 2:of course I mean feature-wise, performance-wise, scale. It ticks every box. The only box it doesn't tick is sovereignty and people are worried about it.
Speaker 1:That's right. Well, by definition, right. The cloud is supposed to be just the cloud, Like. I don't worry about you, know yeah.
Speaker 2:And it's a great service. But the only box they don't tick and that's going to be a very important box is the sovereignty. And of course, green is another thing environmentally friendly infrastructure. But this data processing, it's not just about privacy, it's also about control over owning access and having full access and control over the data. And if you talk about privacy, it goes beyond hosting it in Europe, because legally there should not even be a US link in there. That's quite heavy. And it goes further you don't even allow non-European staff to operate that infrastructure, because they would have access to that data and you have to do stuff like data anonymization and data minimization.
Speaker 2:So those are all kinds of efforts that we've been doing with our cloud. So that's cool, but I must say that this sovereignty is picking up right now. Those are all kinds of efforts that we've been doing with our cloud, so that's cool, and but I must say that this, this, this sovereignty is picking up right now. We're getting customers calling us and say hey, we're doing all kinds of video hosting services for, uh, medical, uh customers. We need to move to a european sovereign facility and then that's that's something we see, really see picking up right now.
Speaker 1:We even are contacted by politicians and they're like please inform us what are the problems we're facing and what are the solutions we can have how to solve it, because they need answers, they need solutions, you need alternatives as well, but the thing that's really picking up right now is on-prem, yeah, yeah, but the thing that's really picking up right now is on-prem, yeah, so that's really popular right now.
Speaker 2:That's not even something we have to sell. People are calling us and say we need an on-prem solution for Windows coding and packaging. Yeah, yeah, yeah.
Speaker 1:So we're seeing it too and I really want to talk about. You know what you built. Also, you know I'd like to hear from you, and I think our listeners would really appreciate as well, because everybody, the market largely flipped, you know, from. It's hard to believe, but you know, 10 years ago for a lot of workflows it was only on print right and then all of a sudden the cloud came. You know the cloud is here and so you know you'd go to IBC, go to NAB, go to any video engineering conference. You know the sessions were jam-packed, standing room only that were about the cloud. You know everybody was trying to understand. How do we migrate? What is it? What are the ramifications? What do I need to know? Technology is supported For many of us.
Speaker 1:Again, speaking as someone who's very much in the middle of the industry, we then rebuilt or migrated on the cloud and then one day you wake up and you say my cloud bill is out of control and the problem is is that you get no economies of scale and so it's great when you're small and you say, hey, this is awesome, I only pay for what I use. You know well that was needed 15 years ago when server cost and infrastructure cost was a million dollars just to get started. Just to get started, I mean, I don't even have customer one. I got to invest a million dollars. So along comes your favorite cloud provider and I can register a credit card. This is incredible. Well, fast forward. 6, 8 forward, 6, 8, 10, 12, 15 years later now the video business is super robust, as you said. You know we're streaming 4k, we're, you know we're at least hd at very high resolution. Using advanced codecs requires more compute. And then I wake up every month and I say, look at my Like this is not sustainable.
Speaker 2:Well, the funny thing is technically, of course, it scales beautifully the cloud.
Speaker 1:Well, technically, but economically.
Speaker 2:But financially there is a challenge. But it's not just the cloud invoice, it's also your own operation. I mean, the promise of the cloud was that your internal IT operation cost would go down by half and that's not true. It's proven to not be true. So operational costs are at the same level or even higher by using a cloud and the cloud recurring costs are extreme. So that's not a driving factor for customers to go on-prem and of course. But an on-prem solution should be scalable, should be affordable and should also be maintainable.
Speaker 2:So going back to standalone applications, like I have this box with a piece of software running on that box, that's not the future. It needs to have the same scalability as a cloud, but then maybe a micro cloud, not a massive cloud, but a purpose-built cloud, and that's what we tried to build with Maelstrom. We were contacted by a customer and they said we have this specific use case and we need video transcoding on-site in a very local facility. We downstream like 200 channels through satellite. It needs to be transcoded to adaptive bitrate streams, push out to HLS and Dash on a local network. So we took this cloud philosophy like scalability. So the fundament of the solution is this Intel or ARM server with those NetIn accelerator cards, but instead of putting on one software application that is running everything, but instead of putting on one software application that is running everything, it's a micro cloud stack. So we build our own cloud software into that machine and it's running a micro cloud stack, which means that every transcoding process is in its own container and if it crashes it's automatically restarted. So you don't have to worry about this and it doesn't crash everything. So there's just one channel, right, instead of taking down everything. And that's the beauty of the system.
Speaker 2:And if you stack multiple servers on top of each other, it becomes this micro cloud with extreme high availability. So if a machine dies, the software auto redistribute all the channels, all the processes across the other machines and you also don't have to worry about anymore how to provision a channel. Like I have so much capacity on this machine, I have to provision this channel on that machine. No, that's not necessary. We just provision channels and the system will figure it out and load balancing for you.
Speaker 2:So suddenly you have scale, automated scale, with all the cloud benefits on an on-prem solution you have the economy benefit of. You know it scales financially because those servers are relatively dirt cheap, especially with acceleration, because you know you can run something at 500 watts what used to be like 5,000 watts on a location. You don't need a rack full of servers, you can just have three the three you with three machines for high availability solution, with three machines for high availability solution, and then operationally. So again you have the technical scale of the cloud, financial scale, but also operationally, you just provision the channels, you choose whatever bitrate that is you'd like to have, you start the channels and it's running, so you don't have to manage and monitor it on a daily basis. So it's got all the benefits of the cloud plus the benefits of the cost and management experts in there. So, yeah, we actually won some deals. Now, One was a very cool deal with a large Dutch European OTT provider.
Speaker 1:I can't name it right now. That's amazing yeah.
Speaker 2:But we won it from Atam, from Amazing, From what's the otheral, from Harmonic. So those were competitors and we won it because of this cloud stack and because of this green and efficient way of running Interesting.
Speaker 1:Well, I would love to hear more, because what you just described is often summarized loosely in what people talk about hybrid cloud. You know, a lot of times you'll hear vendors even pitch. You know, we have a hybrid cloud solution you know, Doesn't have AI as well? Yeah, and some.
Speaker 2:AI functionality.
Speaker 1:Yeah, yeah, that's right. Sprinkle a little AI in there and a little bit of everything else. You know, all joking aside, it is true that the market has certainly identified that there are both technical. There's some very technically valid reasons why you would want to run services in the cloud. There's also technically valid reasons why you would want, or even, as you're pointing out, even all the way down to like well, the laws require it. You know that you run on premise, but there's increasingly the need to be able to sort of flex. You know, oftentimes for capacity need, be able to sort of flex. You know, oftentimes for capacity need. You know I don't want to have a whole bunch of machines sitting in a data center. That 90% of the time are unused, you know, but only for the Olympics. You know, every two years do I turn them on, like that makes no sense, right? Yeah?
Speaker 2:It reminds me of how we built our multi-CDN stack like 10 years ago.
Speaker 1:Exactly, exactly, and the barrier, absolutely. It's so interesting you talk about multi-CDN because now I think multi-CDN is more discussed in the almost exclusively. Well, this particular network has better coverage in this region of the world, this one has better coverage in these countries. But it used to be. It was capacity right, it was just sheer capacity like this one platform can't handle at peak times, so we need to be able to.
Speaker 2:you know, flex yeah we built our own micro cdn in europe and we said we're not going to compete against, you know, the large cdn vendors out there why would we we do that.
Speaker 1:We will never do that.
Speaker 2:But we want to have our own playhead capacity and we build capacity on our regular capacity needs. So just build a stack of caches that can handle the normal load that we have plus some headspace. But for all the overflow, peak traffic we just push everything to the CDNs because then you can have temporarily huge capacity on top of it. You can stack multiple CDNs, because then you know you can have temporarily a huge capacity on top of what I mean. You can stack multiple CDNs on top of each other, on top of your own network and that's still running as we built it 10, 12 years ago.
Speaker 1:Yeah, yeah, amazing, that's amazing. I would like to go back to this project that you referenced, the Dutch media company, and I know you can't share the name and probably you know you can't be too explicit on the exact specifics, but at a high level, can you characterize what they had before, the problem they were trying to solve and how you solved the problem you know? So I think that journey would be of great interest to our listeners.
Speaker 2:Okay, the old setup was not a redundant setup. So in the case of an outage, they had outages Not that many, but the risk of having an outage was a problem. But the cost of running a fully redundant setup I mean it's a complete replication of the same setup running at the same time it's expensive. I mean, if you have to transcode a lot of channels it's going to be an expensive operation. So there was no business case in having a redundant setup. But they wanted to have a redundant setup. So that was a challenge for them.
Speaker 2:We offered a redundant setup which was so much more efficient like 90% less energy consumption in just a few machines on a zone that two zones was even cheaper than a single zone in the old solution. So that was the first thing. The second thing is they also wanted to offer their customers a more economic, friendly but also I've said it a greener solution. So they can claim yeah, we're focusing on the environment as well, something they found important. So that's another thing. I think we won this deal together with a partner company and they also brought in some benefits, which is Dutch, local support, fast support, some innovation, some innovation we do together with them. For instance, we have been doing some extreme tuning to the bitrate ladders and to the packaging to really optimize for buffering and all that stuff. Yeah, exactly.
Speaker 1:And yeah, yeah, sorry for interrupting, but the encoder is the NetEnt Quadra VPU, correct? That's used in these appliances. Yeah, you gave a number out a few minutes ago that I want to bring back up in case somebody missed it, but 500 watts for the entire Transcode server. So let that sink in. Because now for this particular project, did you have eight NetEnt VPUs in there, or do you recall exactly how many cards were in?
Speaker 2:that one. The setup is at least three servers on a zone and there's two zones, so six machines and each server has, I believe, six of those cards.
Speaker 1:Six cards. Okay, yeah, exactly, I mean it's based on physical configuration and also just what the needs are. We generally see anywhere from somewhere around 5 to 6 to as high as 10. And obviously, with the larger boxes you could have even more. It's not uncommon.
Speaker 2:There are some bottlenecks. It's not the cards, it's the machines themselves.
Speaker 1:Correct, correct.
Speaker 2:We took a server, filled it out with a number of cards and then simulated as many live streams as we could to see where the bottlenecks are. The bottlenecks is not on the cards. Well, basically, eventually you run out of card space. But there are other bottlenecks in the machine, mainly due to like if you need to do some deinterlacing or other processing that needs to be done on the CPU. And then you need to have some scalable architecture in those machines as well.
Speaker 2:But we're safe to say that on a single standard box with running on 500 watts we can process easily 30 full HD channels simultaneously with like six or eight bit rates in a letter. Yeah.
Speaker 1:Yeah. So you know I love it and I love that data stuff because, again, you know it's easy to sort of. You know greenwash the environment, it's easy to sort of. You know greenwash the environment and I, you know I see a lot of claims in the market about, you know, products and services, platforms that are net zero and environmentally friendly, and then you just look at how they're architected and you say well, it's not even possible.
Speaker 2:what your claim is, you know you're doing everything on CPU.
Speaker 1:What your claim is You're doing everything on CPU. You're running the most advanced codecs, like AV1. You're trying to run it live and you're saying that you're running a net zero operation. It's not possible. So we really love to know when our partners are truly contributing. You know, to um, you know to green operations.
Speaker 2:The greening of streaming yeah, yeah, yeah, so yeah, but it's, it's real proof, it's real life proof of what we did. We really, you know, we took such a machine, we overloaded it with channels and see how far we could take it and I think at 60 channels the CPU ran out of capacity.
Speaker 1:Yeah, but that is with full ABR ladders, like you say. That's even including deinterlacing. So now are these x86 or ARM-based servers? For this particular, these are ARM.
Speaker 2:So it's like a 96-core ARM machine, yeah, and that's Amp ampere. Yeah, yeah, it's. It's great stuff, uh, because the energy consumption of these arm machines are even lower than the intel machine. Yeah, yeah so we we really max it out and over, try to overload it and see how far we could get it, and that's that gives us better proof also towards the customer. Say, yeah, we're safe that you can run so many channels on a box.
Speaker 1:Yeah, by the way, we have a white paper coming out with Ampere shortly, it should be, so we're taping this the first week of March I think by the end of March by NAB that will be out and it's, you know, basically a technical explanation of how we can achieve such amazing density deinterlacing performance in software on the CPU. You know, when you've got a bunch of ARM cores, you know, when you have 96 highly efficient compute cores, you can do a lot of decoding and deinterlacing and you know.
Speaker 2:Yeah, the scaling is done on the card, so that's also helpful, right, and the encoding, which is the heaviest part, is also done on the card. Yeah, exactly. For the rest, in terms of memory or networking bandwidth, those machines are basically doing nothing.
Speaker 1:Yeah, it's very easy and you are correct. You know. You also referenced a couple of minutes ago that these are not heavy-duty, super expensive servers. They're not the lowest cost, certainly, that you can buy, but they're very middle range.
Speaker 2:I think performance versus price ratio is….
Speaker 1:It's through the roof.
Speaker 2:It's very good and also the bit rate optimization was also really cool. You know, netflix has this, uh, this cool algorithm. They developed it themselves. Yeah, go to uh, you know, to keep maximum uh image quality, visual image quality, uh, and then then you can start optimizing the bit rates. So we optimize every bit rate to achieve the highest image quality at the lowest bit rate.
Speaker 2:So yeah, it's not just reducing the the bit rates, but if you look at how many viewers there are and the traffic that you're generating, so every, every percentage that you can shave off a bit rate is profit for a customer that's right.
Speaker 1:Yeah, that's yeah, that's right now. Um, is it 4k resolution at the top, or or is it 1080p? Or are there a handful of 4K channels? Most are 1080p.
Speaker 2:No most of it is full HD Full.
Speaker 1:HD okay.
Speaker 2:Yeah, and it's got a large bitrate layer, which is nice, yeah, yeah, yeah, nice, nice. And what?
Speaker 1:codecs are they using?
Speaker 2:Right now. Yeah, nice, nice. And what?
Speaker 1:codecs are they using right now? It's OK, but obviously you're ready for HEVC and AV1 again, because net end VPUs can you can mix and match. And you know another beautiful thing I know I'm I'm now sort of pitching, but you know it's just important. You know, is it, what's great as well when you use a VPU is it's literally just tick a box and it turns on In software. Now you have to worry about provisioning, like, oh no, I need an entire machine, you know, because the compute complexity you know, to maybe add a couple extra channels in AV1 or even HEVC. And there's another challenge.
Speaker 2:I mean if you need to distribute the encoding of a single channel over multiple servers, you run into all kinds of syncing issues.
Speaker 1:Yes, yes, idr alignments are incredibly difficult, you know, across your bitrate ladder.
Speaker 2:Yeah.
Speaker 1:Yeah, yeah, amazing.
Speaker 2:Yeah.
Speaker 1:Well, it's great. Well, steph, it was really good getting an update. I know this is a little bit of a shorter episode than we've traditionally done, but we're wanting to revisit. We had heard about the great work that you were doing. Certainly excited when you can officially announce this. Customer in the Netherlands and maybe we can have you back. Customer in the Netherlands, and maybe we can have you back. Maybe we could have them come on and you know we could. We could do a little, do a little webinar or something you know, talking about how how the system was built. Yeah, hey, you know, invite them on and say hey, you know, once we get this launched and it's all ready to go, you know, let's come on Voices of Video, and I think there's a lot of broadcasters and media companies throughout Europe and, frankly, throughout the rest of the world that would be very, very interested to learn what you build. So, congratulations, thank you so much. Thank you, yeah, awesome. Well, good, if somebody wants to learn more, where should they go?
Speaker 2:You know about your product and, yeah, we can give a live demo if they want.
Speaker 1:Great, yeah. So go to Jetstreamcom and you know you can also check out the other Voices of Video episodes, the interviews that I did with Steph, and you know we did some live demos there as well. So, all right, well, thanks for joining us, steph. It was great talking with you. Likewise, as always, mark, thank you. All right, well, good, and for all of our listeners, we really do appreciate your time and the attention that you give us. Tune in and you can continue to get more great content about what is happening in the VPU ecosystem, what's happening in streaming media, in the streaming business, in coding. We try to cover it all.
Speaker 2:So thank you again for being a listener on Voices of.
Speaker 1:Video is brought to you by NetInt Technologies. If you are looking for cutting-edge video encoding solutions, check out NetInt's products at netintcom.
