NETINT Technologies about hardware-assisted software solutions - Simon Karbowski & Jan Ozer

NETINT Technologies about hardware-assisted software solutions - Simon Karbowski & Jan Ozer

Show Video

Hi, I'm Jan Ozer I'm here today with Simon  Karbowski, who is the CEO of StreamVX. Hi   Simon. Thanks for joining us. Hi, Jan. Thanks for having me. Can you give us a 30-second  overview of where you're located  and the product and services that you provide? Sure. So, we are a product company, software   product Company from Poland. We're located in  Gdansk, a famous city in the north of Poland. We basically moving appliances toward software  solutions. So what we do on a daily basis we're   trying to accommodate different problems on the  OTT environments like transcoding, packaging,   and delivery networks. We empower the new  revenue models. We're trying to enrich the  

video systems. We are into digital advertising, optimizing the costs, lowering power usage...  We're delivering the critical components into the  systems of different operators and trying to make  our products as scalable as possible. And  we are also moving them toward the new modern  deployments like cloud or on-premise and mixed models. Okay,  

and who is your typical client? How would you describe them and their needs?  Typically I would say that our client will  be the paid-TV operator or the broadcaster,   Telco. That's basically...,  or streaming companies.   It's a... Various parties have anything  in common with video, basically.   What is giving us the common ground here is that we aiming toward those customers to deliver the new technology  toward the ecosystems,   or we want to introduce new technology, we  change the codecs and solve the problems.   The flexibility in our products is giving  a huge advantage to the parties involved. So you call yourself a software company.   What is the typical way that  you deliver your products? So, typically we're delivering this on the  customer's hardware. So, the customer is  

choosing the hardware. As I mentioned before,  this is also a very often and more and more   popular scenario when we're only deploying this on  AWS or Google cloud, or Azure. So, basically, in   the cloud. There are a lot of mixed scenarios or  hybrid scenarios, if you prefer to call them that.  

So that's typical. But of course, we have  different products. Sometimes the products are   allowing to do this as simple as that. And when it  comes to transcoders... So we have a transcoder,   in our case, the product name. It's a little  more sophisticated because we usually need to   advise a little more on the hardware part. And in  this case, as we know, we can do this in the CPU,   we can do this in the GPU, but we can also do this  in ASICs as we do this with NETINT, which we have   a great partnership with right now. So basically,  to quickly answer your question, we recommend the   hardware, and considering that the customer is  saying, let's say, I want to transcode so many   channels, with that kind of ladders, we're trying  to figure out the best approach to cover this   scenario. And we probably want to go a little more  into the details further in our discussion today, 

but an ASIC approach here is  winning in plenty of scenarios.   One of the biggest trends that are pushing encoding decisions in Europe, particularly, is that rising energy costs. What  are you seeing from your prospects   and your customers about energy going forward? and how they select their encoding products? Yes, that's a very, very important factor, as  you said, in Europe right now. We see more and  

more questions. And actually, it's a very, very  precise question. So we're getting straightforward   questions like okay, 'how much on a daily basis'  or how much per channel' will it consume - the   power. So, again, as we know, it's not so  obvious to answer that kind of question because   nobody before was paying so much attention to the  power consumption in video systems. As far and as  

detailed. But the good news is that all as much  as all RFPs are touching that, again, thanks to   cooperation with NETINT, this  kind of answers are coming a   little easier and on the winning side because there are a lot of advantages to this part. Give us an overview. I mean, you can encode in  software, you can encode with GPUs, and you can  

encode with ASICs how do they compare from,  I guess, a watt consumption per stream basis? So, of course, there are some basic differences.  And the density is the one. But that's  it's not coming toward the power... the power answer... So probably  we'll touch it a little bit later.  But if it comes to powers, the NETINT... The NETINT's Quadras andT408s are, using just seven watts of power. And in our benchmarks, it   is very capable. Compared to GPU ..., it's really  like a hundred times much more power on the GPU.

Of course, the GPU itself is doing a lot  of different things which we're not using   in the transcoding. But that's not the  point because, at the end of the day,   the bill comes for the whole device. So here,  the power usage is on the ASIC side, winning   significantly. Because if you're going directly to  the software side and doing this just on the CPU,  

the power is going to be a little better, but  then we have to remember that the density or   the number of streams switching, which we can  do directly in the CPU is really, really low. So, even though, we will get it a  little, maybe better, on the power side,   the amount of the servers, so in total,  the cost of ownership is going to be   completely behind on a comparison scale. Not worth of mentioning, I would say. So, you're saying that it takes many more  computers running to equal the density of a   single computer with multiple T408s, for example.

That's, basically, going to take a lot of CPUs and a lot of servers to do the same job,   which we can do on one ASIC. So swapping ASIC for servers is  a real win-win for your customers   who are looking to cut their power costs? That's exactly the point. What we, basically, are  proving in more and more cases right now is that,   because basically, VX transcoder is a  software that is leveraging different   platforms for doing the transcoding  tasks. And, of course, as I mentioned,  

we have a different scenario trained and checked.  And we have different, let's say, deployments,   but then ASICs come with really unusual density, I would say. It's, of course, because the ASIC  is dedicated, as the name states. It's dedicated for a particular task. And  as we said, CPU is generally for everything,   so it's not dense enough. It's it's capable  to do a lot of stuff but not only transcoding   or encoding. And then the GPU is pretty  capable on the encoding, though not as  

much compared to the ASIC, but then it's  doing a lot of stuff and it's consuming   a lot of power. So nowadays, I would say,  that is not necessarily the best choice. Do you have any customer success  stories that you can share? of saving power by switching to ASICs? We basically having, right now, or participating right now in the several RFPs and several POCs with the customers,   and as I mentioned before, the winning  trend here is for the combined solution.  So, StreamVX software on top of NETINT's ASICs is basically beating everybody on   density and on power usage. So, what we  see already as really impressive numbers,  

we can squeeze like 30 channels out of one RU  unit compared to a similar configuration on two   RU units. So you already have like half of the  rack space Coming to power usage - it is really a   significant difference. Because one RU unit we're  speaking about, the server is using barely 7,700   Watts. And the two RU units are using 1,500. So,  twice more, basically. And the density is at least 30 percent more channels out of half of the  space. So, let's say KPIs here are impressive,   and at the end of the day, the simple scenario,   where we basically can do this as almost  'Plug and Play'. That makes it even better. I mean, it's nice to hear about power,  but the quality is always critical for   any service. One of the older knocks  about hardware encoders was that the  

quality couldn't match the software.  What's been your experience there? On one side, we have a lot of movement  in the video market. We know, and we're   speaking very often together about that it's moving very fast, and the amount of   video is rising rapidly, and it's going to be  high and higher every day. But the quality,   in my opinion, is getting toward the point  where it's one of the major differentiators   on the market. So if you really want to push  your service further, or you want to gain more  traffic on your platform, and gain more customers,  quality is the factor that will differentiate you. And when it comes to NETINT, the gain in quality is something   that we got quite impressed about. Because, okay,  if we can squeeze it on the software transcoding,  

we can go pretty high with the quality but again,  the density is really suffering here. And on the   NETINT, we see very, very often we see two factors  that get very important, let's say, subjective   impression on the quality level. One is  ultra-low latency which we can squeeze with   the hardware on ASICs, which is very important  for sports or live events. Because then you   want to be really, really fast and just after the actual camera over the source feed. But on   offline, when we have a little more time  and better settings for the transcoding,   then we see VMOV scores like 96, 98 even  sometimes, which is not so easy to squeeze out   of the software. But even if you do that, you're  basically doing just one movie per one machine,   which is useless because it's wasting a lot of  power and a lot of resources. So the quality  

here is, in my opinion, very important.  But also NETINT is quite impressive on   that factors. You mentioned latency what how  many of your customers care about latency? You know it's always about the use cases. For us, every customer is a different use   case. Our flexibility here wins a lot of  cases because we really understand the video,   as you guys understand the video.  It's a very strange animal. So  

you have to feel if what you measure really matters or not. There are some   customers, as we spoke in the second before, which need live sports or gaming events   transmitted. In that kind of thing, the latency  is critical. And there are some customers that   just have a video libraries and the latency is  not so critical. It's just offline stuff, so then the quality is probably the bigger  priority. So it depends. But again, here, as I  

mentioned before, latency is counted a little as one of the quality factors, in our opinion. Our   observation in the market is the latency  is also considered a part of the quality,   and with NETINT, we can squeeze something around  10 milliseconds, which is really also impressive.   Very useful for live events. Definitely very useful. And quite   impressive again comparing to the software solutions. Put that in perspective for us. I mean, if I'm  watching a soccer match on TV what's my latency  

going to be if I'm watching over cable and what's  it going to be if I'm watching via streaming, say,   with one of your systems? It's,  you know, a very good question.  Very hard to answer precisely because every  single ecosystem we see on the market and in   our customers' cases is really different. So  those architectures are, basically, you know,   hard to compare. But the good point here  for you is that we see a huge difference in   the transcoding platforms which are deployed  in the legacy systems. Like a lot of cables  are using steel right now on the broadcast,  where we see even 20 or 25 seconds delay   towards the actual feed. And again, on the  systems where we're just on the transcending,   we're getting 10 milliseconds delay even if  you add, let's say, I don't know, another that   towards the distribution, we are still, you  know, far away from this 20 seconds. So that's,  

in our opinion, it's a huge difference, and it's  a huge jump toward being really closer to the   live feed. So it's a significant difference.  But, again, there are, of course, and I'm not   saying there aren't a lot of deployments that  are having this differently applied, and it's   getting closer, but definitely a basic approach  from NETINT, this is making things much easier,   because that's a critical part which is on  the hardware processing of the source feeds.   It is narrowed towards really impressive  times. Talk to me about integration. So,   if I have a system that's depending upon  software encoding at this point, and I want   to switch to ASICs what's the integration task  look like, and how long is that going to take? That's probably the whole thing  behind our partnership with NETINT And the whole trick here is that  we are a software company. Yet,   we've been allowed, thanks  to this great cooperation, to integrate very deeply with the ASICs. 

So what we did we basically hid all  the magic and all the heavy lifting which is done by the ASICs, in behind, of course to do this fast and good quality. Great  quality transcoding in the back. We've   hidden this in in the nice UI and the nice API, if you prefer to use API, in a nice API to control   this with a few clicks, and easy and fast.  So, from the perspective of the customer,   if you're speaking really, really easy  case, we're just getting some inputs,   and we're just setting up some outputs, and  that's it. So it's going to be a few hours,   basically the simplest possible case.  If we're doing some retrofits or we're  

changing the architecture, we're speaking about  maybe more sophisticated redundancy scenarios,   it's going to be longer. But the whole thing  here is that it's a software-defined solution   with really well-integrated hardware. So this part is completely behind   everything and the customer doesn't care about that. The whole integration is

about, let's say, architecture and fitting in the  stores, the needs of the customer. And that's of   course our job. That's what we do. But that's,  again, case by case approach. What is critical   here, I would say, is that our approach and  NETINT's approach are very similar. That's   why we really like this cooperation because we  really focused on solving the problems. We really  

focus on the customer cases. So these combined  powers are giving us really the flexibility,   so we can say it's a plug and play, Because,  we basically can deliver something which just   giving us this input, giving us the needs  on the output and we will simply deliver it.  So the customer sees nothing but lower power  bills and higher density, and you handle the rest? Yes, that's basically the whole thing here. What's important here... We have to  remember that, and we both know that, the   whole fuss behind the how it's  happened and what exact parameters you have to add to that. You can mess up in 

plenty of places, but in most of the  cases, if we hide this whole thing behind the nice UI, we cover 99  percent of the cases, and then  if customers need more expertise, that's where  we come. We're fixing this, we're fine-tuning,   we're advising, or, as we spoke, deploying  very sophisticated redundancy scenarios   to just make sure it'll never fail. So, what other hardware do you  consider in a potential deployment?  Say, moving from software-only  to hardware-assisted? We have different experiences, and we're doing  this for quite some time already. So we, of  

course, again, as a software solution, we can run  it on the CPU, we can run it on the GPU, and we   can run it on the ASICs. We integrated different  hardware accelerations. But most of the cases,   which we see right now, again, which we  mentioned already, are taking more and   more care and are more interested in the power  usage and density. For obvious reasons. And as   much as right now is the case in Europe, because  we know the crisis situation, all the stuff, but,   in my opinion, it's going to evolve it's going to  be very important globally very soon. Because the   rise in the amount of the video which is produced,  and so on. And so it will simply force everybody  

to take care of the density and the power  bills. So what happens, and what we see, is   that the ASICs advantage ahead of GPU, is really  significant. So, we can really go like a 10, 20,   even 30 times sometimes the density with the  much, much lower usage. So, if you're counting  

this venture like a TCO three years, it's really  easily covering the whole cost of the difference  through the power bills. It's really so  significant. And, of course, we can run a   different acceleration but what we see right  now is really migration towards efficiency. So, we probably even gonna see soon retrofit  cases like somebody we're gonna want to exchange   some hardware for a more efficient one. Which again in our software case is   very very easy. And similar to NETINT, which is based on the interfaces which   can match probably to virtually any server  on the market, because the interfaces you   can use for connecting the hardware towards  the server is universal. As our software  

can universally run in a lot of servers. So  it's, again, a perfect match. We can basically   even rescue a little older systems  which they want to optimize. You've mentioned the payback period.  What are you seeing is the payback  period for a T408 type installation? That's, of course, coming  to the scale question again.  But what are we doing right now n the RFPs, in the cases we right now working on,   it's usually three years which easily  excuses the cost. So that's not too long,  

I would say, but it depends on the scale.  There are probably a lot of cases that are   going to repay it faster. There are some  cases in which, maybe, big installations,   or recent installations, which are maybe  not so power-consuming. But in general...   It's not any more discussion that, you know, you're really looking for a single person.  

With the density gains, we're getting on this  kind of approach. It's really easy to catch the   moment you're getting back the money. And nobody's predicting that the  power pricing is going to go down   in the near term? It looks like  it's always going to be going up? Probably... If you both know that, we would  be rich or we wouldn't be sitting here.  Nobody knows. Honestly, nobody  was expecting the pandemic 

 or the war in Ukraine. So  it's not so easy to answer, if it is definitely going to happen like that  or not. What we can be sure of, in my opinion,   that's, and we see this for years, and it's giving  more traction right now, that amount of the video   produced and the live video produced right now  is rising rapidly. And because the amount of  

the video content is rising rapidly all the  time, so the amount of the infrastructure   which is processing the video also needs to be  raised. So definitely, this cost is going to be   taken into account more and more. And we have to  remember that not only the live video is here,   but there is a lot of offline content that is  there and must be processed somehow. There is a   lot of gaming content that is being streamed,  okay. That's also live, but it's a little   different animal. There are a lot of advertising  markets which is needing transcoding as well. So  

there are a lot of different places where  this video processing is very important. So   even if the prices were going to stay, the scale  is going to rise, and the costs are going to be   brought to the attention, and the cost gonna  be kept low by density again, in my opinion. So summarize that, you know, if I'm  out looking for the encoding hardware,  you know, we've talked about...

We've talked about costs, we've talked  about the power, we've talked about latency,   we've talked about the quality, we've talked about  the interface... Are there any other factors that  potential buyers of encoding  products should consider? A lot of them! But I have to mention  a few.I really like, you know,   with the cooperation we've got here. Because the video itself is really an  

ugly animal, So what is very, very important in the environment we're working with is   really the awareness of what the video is,  and how to behave with the video. And again,   StreamVX is coming with a lot of history  in the industry, and NETINT with a lot of   experience. This is a perfect combination to deliver a partnership that will create a   video venture in the right place. Because it's  really really important to understand how to   take care of the video itself. The other part is  all this quality stuff, which we spoke briefly,  

that it really there is a lot of video on the  market again, but the quality is starting to   be a differentiator. In my opinion, it's  really going to be huge. Because, okay,   we're speaking about the 4K we're speaking about  the 8K, but we're still seeing really ugly even   HD. But it's not HD it's not even close to the HD. So, we're still seeing a lot of streams that   claim to be HD, but they are far away from the  quality we're expecting from the HD streams.  

And this is also very important to remember  that does not necessarily have to be like that.   And it shouldn't be like that. So that's  another factor we have to take care of.   And very important is this flexibility. The  flexibility of the vendors, of the solution,  

or of the product that you're choosing for  the transcoding or encoding. Because we have   to remember that the use cases or the number of  streams you want to process are changing. And the   conditions are changing, our inputs are changing.  And the flexibility of the product is very,   very important because we see right now, in our  experience, that there are a lot of appliances,   which are, as usually appliances are, built to do one function. And we're living in the   times when the software approach for the function  changes with the update of the software, and it's   as easy as it can be. Just click, and here you go.  That's something that customers expect right now.  

And on top of that, if you want to make really  sophisticated or higher ability scenarios, where   you don't want to allow it to fail on the live  event, or you don't want to allow the VOD system   failure on the, I don't know, dozens or thousands  of customers. You want to be sure that this   solution you choose is really capable to do that.  And again, the software-defined proper design as   the StreamVX, with the proper hardware support,  which gonna be NETINT, you know, you are always   on time and with low latency and good quality. That's really a perfect match, in my opinion.

Well, that sounds like a perfect  ending to this interview.  I appreciate your time,  Simon, and talk to you soon.

2022-11-08 22:53

Show Video

Other news