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
How Scalstream Replaced 40 Servers with Just Two
Transforming video processing with groundbreaking efficiency isn't just a tech achievement - it's an economic and environmental imperative. Scalstrm's Dominique Vosters joins us to reveal how their innovative approach to origin packaging and transcoding is delivering extraordinary results for broadcasters and video providers worldwide.
When a single customer replaces 40 servers with just two while saving $90,000 annually on operational costs, the industry takes notice. Founded in 2017, Scalstrm built their high-performance video processing solutions from scratch, focusing on three core principles: performance, flexibility, and ease of use. Now, they're bringing that same philosophy to transcoding through a partnership with NETINT Technologies.
The conversation explores how just-in-time transcoding is finally becoming viable at scale through hardware acceleration. By storing only the highest quality profile and generating lower resolutions on demand, Scalstrm helps customers reduce storage requirements by 50-60%. Their intelligent approach to Multi-Profile Video Recording (MPVR) keeps all profiles for the first few days when viewership is highest, then removes lower profiles to optimize both performance and storage efficiency.
We dive deep into deployment scenarios across on-premises, private cloud, and public cloud environments. The emerging hybrid cloud model enables broadcasters to run core channels in their data centers while quickly spinning up "pop-up channels" for special events in the cloud. With Scalstrm's microservices architecture, customers can deploy exactly what they need, where they need it - whether that's live processing in one region or VOD in another.
Sustainability emerges as a central theme throughout our discussion. Beyond the marketing buzz, Scalstrm demonstrates how technological innovation directly contributes to both cost reduction and environmental responsibility. Visit the NETINT booth at NAB to see Scalstrm's just-in-time transcoding solution in action and learn how your organization can achieve greater efficiency in video processing.
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 video voices of video welcome back to the special edition of voices of video. I am so excited to be doing these short interview segments with a number of our partners that we are going to be featured both in our booth as well as in other various settings around NAB. Today I am meeting with ScaleStream and so, dominic, first of all, thank you for joining Voices of Video. Welcome.
Speaker 2:Yeah, welcome. Glad to join the sessions.
Speaker 1:Yeah, absolutely. You know, we were talking before we hit record and there's so much innovation happening in the industry and our two companies are doing some really great work together. Our two companies are doing some really great work together. So, first of all, why don't you introduce yourself, tell us about ScaleStream, and then let's dive into a conversation about how we're working together.
Speaker 2:Yeah, definitely so. My name is Dominik Fosters from ScaleStream. I'm responsible for business development and sales, although I have a quite large technical background, so that's why I'm also involved on the sidelines in benchmark tests with NetInt and so on. In the beginning, also involved in improved concept and so, but lately, mainly focused on business development and sales, especially because, since the company Skillstream grew quite a lot with more people, now we have a key focus on everybody on this job and minus business development and sales, the company Skillstream. Perhaps a short update for the people who don't know Skillstream yet we are based in Sweden.
Speaker 2:The company was founded in 2017. Initially, the origin was our origin. Slash repackagers was our main product. This is a product that's developed completely from scratch, so we are not using any open source components glued together with scripts. Everything is built from scratch with three things in mind Performance, ease of use and flexibility, and later on, we also added Textra products. As you can see on the slide, we have a Shield, a CDN, i2t, image-to-text subtitle conversion and lately transcoding. And this is why we are talking, especially because, with the NetInt cards, this allowed us to keep the performance high, which is quite high on our list sustainability. On the origin, for example. We are 10 times more performant compared to competition At some customers we replay replaced 40 servers with just two, and that's what we want to achieve with transcoding as well.
Speaker 1:Yeah, it's amazing 40 servers and you replaced all those with only two. Yeah, that's a lot of efficiency, that's incredible. I don't know anybody who isn't interested in two things cost reduction, and usually massive cost reduction. And then it goes hand in hand with energy reduction. Right, absolutely.
Speaker 2:Yeah, yeah, no because I one of one of our customers, did a case study by replacing so many servers and they saved around was it 85 or 95K yearly, mainly on power supply operational costs. So sustainability is really high on our agenda and we can talk about it in more detail. But we believe we're convinced that we can do the same on the transcoding side as well, together with NetInt.
Speaker 1:Yeah, absolutely Well. Thank you for that introduction and overview. I'm actually really interested. And I think where I'd like to start, I know you have a couple slides prepared. Listeners, don't worry, we're not going to bore you with a couple slides prepared. You know, listeners, don't worry, we're not going to bore you with a PowerPoint, but you know, I think you know they help frame up the information nicely. But I would like to start with. You know there's some very interesting trends that are happening and one of them is is it the function of the transcoder is transforming, you know, might be the, you know, might be the best way to put it. It used to be. Transcoder is pretty simple you took a file in, you took a bit stream in, you output, you know, some assortment of resolutions and you know, and then a product, like your original product, the packager would take that, package it up and make it, you know, available for streaming, right, um?
Speaker 1:now it, it. It seems like a lot of these uh functions are kind of, are kind of merging or they're sort of blending together such that, you know, the transcoder is almost as much of a router you know, a signal router as it is literally a file transformer, and so maybe you can give us the perspective from your customer base. You know how is the, you know how is. How do you guys think about this? As you have an origin product, you have your shield, you've got the packager, you have. You know, that's a part of the transcoding solution.
Speaker 2:It's a kind of drop-in component on the origin, so it's just kind of plug and play. You install the right microservice and you're ready to go. It's fully integrated in the existing product, in the UI. It does auto-scaling in the cloud. It distributes the load among multiple servers if you have multiple servers. So that's quite a good add-on to the product.
Speaker 2:And, as you mentioned in the past, it was, say, offline VOD transcoding. Of course we support VOD, offline transcoding, but what we saw from our customers is cost reduction. This is very important, more important than ever, I think. And yeah, therefore we also added some new products in the transcoding space. I know that that's really the only one, but like, for example, just-in-time transcoding for VOD.
Speaker 2:We did some calculations based on customers and they can save roughly 50% to 60% of the storage, meaning also 50% 60% of the storage cost. And the way how we do that is mainly, for example, for VOT or also for MPVR. We only keep the highest profile on the storage and we repackage the ABR profiles on the fly. So, yeah, the highest profile takes off, often the highest storage space. So in that way we can save 50 to 60% of storage. And then sometimes we get a question from customers like yeah, okay, you save on storage, but don't you lose that?
Speaker 2:On transcoding power, and two things. First of all, with the NetInt cards we can talk about figures later but first of all with the NetInt cards we can achieve quite a high performance, transcoding performance, at a low cost, and that's one. And also we do it in a kind of efficient way. For example, for MPVR, typically the first 12, sometimes even 36 hours are the most popular. So we keep all profiles for two or three days, whatever we configure, and only after three days then we remove the lowest profiles. So in that way we don't lose performance or transcoding power compared to storage space. So these are two optimizations we've done in that space.
Speaker 1:Yeah, amazing. Now are most of your clients deployed in public cloud. Is it in cloud, but their own cloud, their own data centers, or is it all on-prem? What does the architecture look like?
Speaker 2:Well, it's a bit of a combination. What we see at the large telcos typically a lot is on-prem, although we see a clear shift towards public clouds. So this is definitely a shift. We see Smaller broadcasters typically run fully in cloud, so that's why it is quite easy to spin up. So the skills stream solution as such is microserve, so we can spin it up in on-prem Kubernetes, public cloud, private cloud. But the same is valid for transcoding. And these days, if you look at the most popular cloud vendors, they also have NetInt cards available for selection and this makes it quite powerful that we can just spin up instances. If they need more power, we just spin up an extra instance, kind of auto-scaling in the cloud.
Speaker 1:Yeah, I assume you're referencing for example, like Akamai on Linode. Yeah, exactly so. So this is really fascinating. You know the highest profile, you know sort of the quote-unquote mez file, if you will, and then creating derivative files literally on the fly. There's tremendous, you know, from a usability perspective. You know, you know from a usability perspective. You know you think about now I could literally have, although I would never do this, but in theory you could have perfectly matched profiles on a device by device basis. You know absolutely matching one for one screen sizes, because all it is is a config file.
Speaker 1:You know, in normal, in normal topologies, well, you'd have to go encode that file, you'd have to store it somewhere and then, who knows, maybe just a very, very, very small subset of users have that particular device with kind of an unusual, you know resolution or something unusual about the format support. But this whole holy grail concept of just-in-time also broke down when you had to do it, cpu software on CPU because it was just too expensive. Or to get the bit rate, efficiency and the quality. You could run the encoder super fast but your bit rate was going to be really high. So you lose your efficiency, your codec compactness, but then you also generally lose quality. So talk me through. What was your journey to go from CPU to VPU, for example? Did you look at GPU? Did you even deploy GPU, or did you just jump all the way from CPU and adopt VPU?
Speaker 2:Yeah, well, initially, our offline VOD transcoding solution. This was the first step into transcoding business for us. This was initially running on CPU. It runs fine, but, yeah, it needs quite some CPU resources. Of course it's offline so we can say will you schedule the jobs? And it can be done. If it's done in an hour or in two hours, okay, it's not critical. Of course. If you go into just-in-time transcoding, then it's a different story. Then you need to have the power to deliver it in a matter of milliseconds or a really short time frame.
Speaker 2:Yeah, and then building this based on CPU, there we saw a bottleneck. If only one channel or a limited number of assets, all fine. But of course, if you go into the tier ones or the larger broadcasters with a lot of content, yeah, then the use case falls apart, more or less, especially cost-wise. And this is when we start searching for other solutions and we also looked at GPU. But then, of course, with the net, vpu cards, the quad, yeah, we did some tests in the initial tests and then this showed, yeah, a huge increase in performance.
Speaker 2:Um, so we kind of uh, the balance is perhaps not the right word, but the focus is mainly on, uh, on the net in cards, for sure. And this is also when we we decided, okay, let's, let's set up some benchmark tests. We did it. We have some cars on-prem, but to make the comparison a fair comparison, we actually set it up in Akamai Connected Cloud, as you already mentioned, these have the NetInt cards available, because then we can compare exactly the same nodes based on CPU, based on GPU and based on the net-in cards. And then, yeah, we saw the difference in performance clearly.
Speaker 1:Yeah, that's great. What codecs are your customers primarily using? I assume H.264 is ubiquitous, but HEVC AV1?.
Speaker 2:Yeah, av1, not yet. It's mainly H.264 and H.265, hevc. These are the two. There are a lot of talks on AV1 or new codecs coming up, but it's still not widely deployed. I think this will change, probably in the next year or the next two years, I'm quite sure. But yeah, it's still mainly the two projects, regular ones.
Speaker 1:Understand and resolutions. Is it all HD and below, or does it go up to 4K? What are the common resolutions?
Speaker 2:4K content available, but also there it's still mainly HD, and then 4K is not that commonly used, one thing that often is done. We also have on our origin a concept called adaptations, where you can see it as a recipe for the player. We can make an adaptation for a mobile phone, for a smart TV, for a set-top box, and then, based on the device, you can set different codecs, different bit rates, and then typically for the mobile phones the HD is stripped, for the SetterBox it's the opposite, only the higher profiles are sent. And this adaptation also ties into the just-in-time, because also then we can say, okay, we only just need to transcode just-in-time Only these profiles and not all the profiles. So this goes a bit end-to-end, these two features.
Speaker 1:Oh, that's interesting. Okay, and that does make sense because you know there is a core set of files meaning you know resolutions, bit rates that you are going to cover like maybe 100% of your user base, and I'm thinking of some of the lower resolutions and the lower bit rates. There's no need to encode those on the fly. They also don't take up a lot of storage. So is it that you're encoding on the fly, transcoding on the fly, the highest resolution, or I guess it'd be medium resolution, because you're sending over, like the 4K asset right, the top profile it's configurable.
Speaker 2:So the highest profile is, let's say, the input signal or the content, and then we can transcode all the profiles on the fly. But it's configurable. You can say, okay, I only want to transcode the mid-range, let's say, or the low range. It's kind of configurable. But a typical use case is that the highest is used as a source and all the rest is transcoded. Okay.
Speaker 1:Okay, interesting, all right. Well, I know that our listeners are again. You know we're touching on, as I already mentioned just-in-time encoding, transcoding. That's been talked about for really years. It's been theorized as part of the edge, so these concepts are not new and I think everybody you know has certainly heard of them or even done some exploratory POCs to you know to build something out, but very, very few people until now have been able to go into production.
Speaker 1:One of the other things that's been talked about is the idea of the hybrid cloud and the hybrid cloud being where you have on-prem that flexes into some sort of a cloud and that could be your own data center or it could be literally a public cloud or maybe some other. You know colo or you know something With Akamai, with the connected cloud. We are hearing more and more. So my question to you is is this enabling this hybrid cloud type, flexing, you know, to what might be on-prem and then going into Akamai Is that how you're using it, or are you running 100% the infrastructure on Akamai, or what does that look like? Yeah, it's a combination.
Speaker 2:I typically broadcast this. They don't want to hassle around with hardware and on-prem, so it's, most of the times, fully cloud. There are always exceptions that prefer everything on-site. The tier 1 operators, on the other end, they typically prefer to have everything on-prem, although there's a change. But what we also see is, for example, what they do is all the main channels are running on-prem, but pop-up channels, like Champions League finals, olympic Games, special events, ufc and then these special events these are typically spinned up in the cloud, because it's kind of a pop-up channel. It's only. They spin it up for a few days, for a week or whatever, or even for an hour, and then they tear it down again and then, yeah, now that we have this just-in-time transcoding, yeah, this is even better, because then there's only one feed that needs to go into the cloud and then all the rest is done from there.
Speaker 1:So, yeah, this is definitely a valid use case Interesting.
Speaker 2:Okay, this is definitely a valid use case.
Speaker 1:Interesting. Okay, yeah, so what I hear you saying is the use of Akamai or some other cloud platform. It could be in both cases Somebody could be running 100% their workflow on the connected cloud, on Akamai's connected cloud, or they could be flexing into it, running kind of a combination, right?
Speaker 2:A true hybrid. Yes, yeah, okay, yeah, and then from our end it's one UI and then you can just deploy wherever you want. And then you say, okay, if I, for example, I only want to do live in the cloud, you spin up a live instance with the microservices needed for running live. You can have a separate node for VOD only because typically these are different use cases. For example, for video on demand, it's egress based. For live, it's kind of static bitrate that you need and you need to record the local buffer for time-shifted V. So we can kind of mix and match with our microservices and then deploy them in specific cloud nodes where needed, or even specific regions. So if you have a regional broadcaster, for example, or regional channels, we can just stream them to a local cloud instance in Akamai Connected Cloud and then do everything from there.
Speaker 1:Yeah, interesting. Okay, yeah, makes a lot of sense. And, as you said in your introduction, flexibility is very important. Energy conservation I think it all can be summed up in the word efficiency. Right, you know, you've built your solution, you bring to your customers the most efficient, which of course then translates to lower costs, translates to lower energy, translates to, you know, generally easier operation. You know, in other words, it has like a dozen different benefits. But, you know, when you engineer, you engineer.
Speaker 1:It's interesting because we very much approach our product in the same way. I mean, even the very decision to build on an ASIC, using ASIC technology versus some other platform, was all about efficiency. If know, if, if you're going to build hardware that is purpose built for video, then sure you know there's, there's other platforms, you know there's. You know GPU and there's FPGA and there's, you know there's other approaches, but there's nothing like an application, specific, integrated circuit, asic. I think that's why. I think that's why we were such a natural fit for our two solutions.
Speaker 1:Well, good, well, dominic, thank you so much for this overview. Why don't you just, in closing, let everyone know what are you going to be talking about at NAB? And I also would be really curious. You know what are the conversations that you're hoping to have with the industry, with the market, with your customers. In those meetings, you know what is it that's, you know that's important to scale stream but you really think is going to add a lot of value to the industry and so you want to be, you know, talking about those things.
Speaker 2:So one of our main topics at NAB is sustainability. This is definitely a topic that's really relevant for us and actually not for us but for our customers mainly because we don't do it for our own, we do it for our customers mainly. And then at NAB we're showcasing also just-in-time transcoding on the NetInt cards. So this is what we can talk about. We will bring a kind of PC with one of the cards in there that we can showcase. And then, where we want to talk with customers, not only we want to, of course, tell what we can do as a kind of integrated solution with our origin, but we also want to hear from customers what they need, because in the end we need to build what customers need, not what we think we need to build. So I think in two ways. So we definitely want to talk with customers that have ideas and that might need to see a missing feature on our end. We always listen to our customers and adopt accordingly.
Speaker 1:Yeah, that's great. Well, just in case it got lost on any of the listeners, scalestream will be in the NetEnt booth. Do you have your own booth or are you in our booth? So yeah, and do you have another? Do you have your own booth?
Speaker 2:or are you in our booth? Okay? So we are only at the NetAmp booth, Okay?
Speaker 1:great, Great. Well, we're very happy to have you, happy to host you there. So anybody who wants to learn more about ScaleStream, sit down and talk to Dominic and the other team members who will be there, Come to the NetIn booth and they have a nice kiosk with a big TV and he's going to have a demo there showing off the solutions. So please stop by. So, Dominic, thank you again. It was great talking with you. By the way, I think we probably should do a follow-up after NAB and we'll do a little bit of a longer interview we can talk about. You know what you, what you learned, key takeaways from NAB and, you know, maybe there's a few other developments with the product or the company that you'll want to share at that point too.
Speaker 2:So we'll book that for maybe later in April or May? Yeah, sounds good.
Speaker 1:All right, excellent. Well, thank you again for listening to Voices of Video. We really appreciate all of you and look forward to meeting you at NAB.
Speaker 2:This episode of Voices of 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.
