A New Way To Cut IoT Network Data At The Edge

The fastest way to break a modern network isn’t your Netflix download, it’s the quiet, constant upload from sensors, logs, and telemetry that nobody ever reads. We sit down with Julien Dersey from AtomBeam to unpack why data efficiency is suddenly a front-line issue for IoT networking, edge computing, and cloud operations, even in a world with 5G and new satellite options like Starlink. The uncomfortable reality is that bandwidth grows, then data expands to fill it, especially once cybersecurity teams demand near real-time visibility into who connected to what, from where, and when.

We get concrete about the uplink bottleneck that hits IoT deployments first, and why “just filter the data” is a risky workaround. Julian shares a field deployment with an oil and gas fracking operator transmitting over Starlink, where compaction reduced traffic dramatically and kept gigabytes per day flowing reliably for months, while also helping identify odd behavior coming from a sensor. From there, we explore how AtomBeam’s lossless “compaction tunnel” differs from traditional compression, how it can run with extremely low CPU and memory, and why keeping applications unchanged is a big deal for real teams.

We also dig into enterprise and operator integrations: testing with Ericsson over a 5G router and SD-WAN style network bonding, the latency and performance questions engineers always ask, and the security posture using TLS 1.3 with an added obfuscation effect. Finally, we widen the lens to point-of-sale receipt transmission at scale, disaster recovery replication speedups, and what’s coming as connected vehicles, smart meters, and smart grid AMI 2.0 generate even more machine data.

If you care about IoT bandwidth, edge efficiency, secure data transport, and the future of connected devices, subscribe, share this with a colleague, and leave a review. What’s the single noisiest data stream on your network right now?

Everyday AI: Your daily guide to grown with Generative AI
Can't keep up with AI? We've got you. Everyday AI helps you keep up and get ahead.

Listen on: Apple Podcasts   Spotify

Transcript

SPEAKER_00 0:02

Hey everyone, really excited for this chat today as we look at a brand new uh kind of innovative approach to IoT and networking gestion and dealing with exploding data volumes with Adam Beam. Julian, how are you?

SPEAKER_01 0:17

Uh hi, Evan. I'm very well, thank you. Uh and great for having me uh today.

SPEAKER_00 0:24

Well, thanks for being here. Really excited for this chat. And um before we start, maybe an introduction to Atombeam. How do you describe the company and your role?

SPEAKER_01 0:35

Um so yeah, so Atom Beam, uh it's kind of you know, a company that's been around for a little a few years. Um it started with the sort of crazy idea that hey, we could make data smaller. So who hasn't heard about compression? Um, but uh the idea was we can make data smaller in a much more different way than traditional compression. And so that idea, that brainchild, um, you know, translated into, you know, um it's actually was uh backed up by a professor in mathematics uh from South Carolina, who is our chief scientific officer, uh and came up with all the maths and everything. And the key thing about this idea is we can do it with very, very, very little CPU uh requirements and very, very, very little memory. Uh and it's um so I what what happened is uh I used to uh manage all the product portfolio for InmarSat. And um at the time, and you know, the this idea came across my desk, and I went, that's a fantastic idea. Uh, it could work really well for my uh satellite IoT products, and so that's how I get I got to know um Atom Beam. And then you know, um Vasat acquired InmarSat and you know we parted ways. And then Charles said to me, Hey, you know, don't you want to join this startup company? And I had never joined a startup before. I always worked for big companies like Amazon uh for five years, and Madeus uh in the travel IT for 14 years. Um, and so they this came along and um and I went, yeah, why not? Um because the I because the tech was really cool, and so and Charles could see saw what I could do in terms of my ideas around products and things. So that's why I I joined Atom Beam as the chief product officer uh to help shape the product and how we can take it to market.

SPEAKER_00 2:24

Brilliant. And why is data efficiency suddenly such a big issue now? We have bigger pipes, faster pipes, 5G, all kinds of new uh air interfaces. Um why now?

SPEAKER_01 2:36

Yeah, I mean, absolutely. You you you're correct, Evan. So why? Because we have 5G and the data is always uh, you know, so we can we get much bigger pipes. Um reality is uh data is like a gas, you know. If you give it more space, it will occupy the space. So if you if you give people uh bigger pipes, they will find ways to stuff the pipe with more data. Uh and that's what clearly is happening. Um but more concrete examples, um like cybersecurity, for example. So you need to collect a lot of information about what's going on with your devices, who's connecting to what device, what connections have been made, from where to where, etc. All this sort of uh telemetry information from the devices and that happen everywhere. Um, more and more now people realize they need to absorb that information in near real time to know what's going on so that they can take action. That then creates more data being transmitted uh over the network. And by the way, this is data secondary to all the actions that happen from the primary applications. So you now have the primary applications of data plus the secondary, which is how people are making those connections. And is is there a bad actor on my network? Is the network behaving in a different way? That's one example. The other thing is people forget that with 5G and or even satellite or even Starlink, that yes, data uh pipes are bigger, but they tend to get bigger in what we call the downlink, in the download. So, yes, you know, when you watch Netflix, it's fantastic. You can you can watch you know high definition. But what happens in the IoT space, for example, they don't need to download lots of data. In fact, it's the reverse. In the IoT space, you have all these remote sites generating a lot of data, they need to upload the data back to the mothership. And that's where the crunch comes because this 5G or uh satellite Starlink, the uplink is always like a fifth or tenth uh of the downlink. And if you do a speed test on your phone over 5G, you will see your upload isn't much smaller than your download. And that's for good reasons, because as a consumer, you want to download data, but as a business and operating IoT, it's the opposite. So this is where we come in because all this IoT information, all this log information, uh measurements coming from sensors, they tend to actually be quite quite repetitive. And so our system can very quickly learn what are the patterns in the data and replace replace them with something a lot more, a lot smaller that you can't read, but a lot smaller, and then at the other end, we can then deal with it with that information. So what happens is a lot of the data we transmit, we make it human readable, okay? And the reason is because, well, you know, we need humans to read the data, but what you'll find is probably 99.99% or more of the data never gets seen by a human. So it's actually ways to make it human readable. What you want the data to be is you want the data to be yes, computer manageable, um, to be super efficient. But that also will be a very difficult task for engineers and developers to do. And this is where Atom Beam and NUPAC comes in. We can make that task very easy. We'll take, don't worry, you you can send your log information the way you want it, no problem. And then we will take care of making that transmission a lot smaller for you.

SPEAKER_00 6:00

Brilliant approach. Um, so it sounds really interesting, kind of radically different than traditional compression. But what was the first kind of real world use case when you realized personally, wow, this is a big deal?

[Ad] Everyday AI: Your daily guide to grown with Generative AI

SPEAKER_01 6:15

Um, so it one of the the first real world use cases that, you know, yes, we did a few POCs and you know, with some sample things, and that's great. It's all it's pretty much done in the lab. Um, what we did is we worked with the fracking company uh out of Texas data over Starlink. And um we we deployed it and uh very quickly, you know, we were uh reducing the data by uh 80%, so down to the virtual original size, which meant initially they were filtering the data because what happened also with your upling bandwidth as an IoT business is that you need that constant bandwidth available. And the problem with Starlink is during the day a lot of people using the satellite, so you have less bandwidth available because you're sharing it with other people. And so you then and then during the night you have more bandwidth. But as an IT, as an operation, you can't say, yeah, during the day I send less data, and during the night I send more data. That doesn't quite work. And so um, using our software, they were able to send all the data they wanted rather than filtering it down. And in fact, they were considering they could send even more data. And what was quite surprising, and I think they even everyone forgot that the system was still deployed on the on the trial van, and it kept going for about nine months. Uh and it it performed in nine months, sending gigabytes of data every day over Starlink, uh, and with no interruption, no issues. And and it went beyond that because initially, when we first deployed the solution with them, they're starting to say, Oh, we're seeing spikes on the data that drops, and then it goes back up. What's going on? We were able to analyze what we were seeing and tell them actually it's coming from one of your sensors, it's not coming from our solution. So, so there quite a lot of interesting stuff um that we could observe uh thanks to our solution.

SPEAKER_00 8:48

That's fantastic. And I assume uh many of the telcos and cloud hyperscalers, etc., are starting to pay much closer attention to this challenge. What are you seeing, you know, in in partnership with them and and through them?

SPEAKER_01 9:04

Yeah, so I mean, uh absolutely. So we um we we work here we're in partnership with Ericsson. Uh and one of the uh key things is um, in fact, um they we said, okay, what we need to do is let's test our product, you know, does it work over a uh wireless router, a 5G router, uh, and over an SD1? So SD1 typically it allows you to combine multiple network connections and make them look like you're only uh going over one connection. Um and so that's called network bonding, uh, but also it offers capability like uh quality of service, which essentially means prioritizing certain traffic over other ones. And the reason why you want to do that, why do you want to combine two cellular connections? Why do you want to prioritize traffic on your network? Because in both cases, it's because you don't have enough bandwidth. And you know, our product is there to actually help you when you don't have enough bandwidth. But we solve it in a different way from SD1. SD1 solves it by prioritizing traffic, combining multiple connections. We saw we solve the same problem by making your data needing less bandwidth. So the two together multiply each other. And so that's why with Ericsson we we ran that test. And when they tested our product over their 5G router, network binding, etc., um they realized actually um it worked perfectly. And not only that, but you know, everyone worries about how much delay does it introduce, how much latency that adds, how much CPU does it need, how much memory on this edge router. And and the guy said, look, you know, he to find the CPU consumption, he had to look in the decimal points to find uh our product CPU consumption. So that's a testament of how efficient um our solution is. Uh, this kind of thing. We don't do that. We take a bike stream, and all the bikes that go in will come out in exactly the same way. Okay, so it's completely lossless and it it maintains the um the data as originally sent.

SPEAKER_00 11:58

Wow, well, definitely fit for purpose for IoT. And I can understand the performance implications, cost savings as well. What are the security opportunities implications for using this on the network as well?

SPEAKER_01 12:11

So, I mean, our product we call it it's a compaction tunnel. Uh so it's fully secured. We had it pen tested. Uh, so we're using TLS 1.3, uh, we're using the the latest, that's all been verified. So um absolutely no issues there. And you you could argue it had an extra layer of obfuscation because we're actually converting the data into this, what we call these code words, and uh you need to know the code books. So you not only you have to break encryption, but you have to find out what are these magic code words, and they're not like very obvious. Uh and so in fact, the professor in mathematics um, you know, the the theory is the the byte stream that is generated even before encryption already looks like uh almost random noise. It's already maximum, what we call maximum entropy. Um, so so there's no so when you look at it, it would be very hard for you to distinguish where where are the different code words because it's it will look like this random bits, random zeros and ones coming in uh in the flow. Uh so that's that's pretty much um why. But in general, we're fully secure.

SPEAKER_00 13:20

Fantastic. Well, that's uh a nice validation as well. Um you're obviously well suited for almost any industry, but you know, uh some more than others. I guess oil and gas, great example of an IoT use case. Any others that are uh really obvious and you're getting crafted with? Yes.

SPEAKER_01 13:39

Um so um point of sale. So all your credit card devices, credit card machines, etc. Um, and one of the big use cases is every time you make a transaction, um they then the machine stores the receipt that has been printed electronically so that it can then be transmitted to the bank. So if you have a dispute, uh customer says, yeah, you know, it's not what I was charged, etc. Um, instead of requesting the merchant to send a paper receipt and have a photo or something, which is very consumed, very time consuming and you know costs a lot of money, it's not worth doing if you if you're you know claiming for you know chips that you bought at McDonald's, or sorry, I should say French fries, and then you bought a McDonald's, um, then that electronic receipt gets transmitted to the bank. And these things are quite you know, quite large, they're three kilobytes. Um, but you have when you have millions of transactions, so our partner in Saudi Arabia um has been using it for now for almost a year, uh, our product, uh, and is helping them to reduce um you know, on a daily basis to generate five gigabytes, and we reduce that down to less than one. And so that allows them to transmit the information, store it in uh disaster recovery sites much faster, and so on. Um, here's another example where again in Saudi Arabia, um, they we've just been uh discussing with them, they are using it for replicating virtual machines. So all your application servers are running pretty much on a virtual machine, and you want to have a copy of that machine and another site for disaster recovery purposes. And we are using they are using our product to make that copy up to two times faster or even two and a half times faster, uh, which is quite unheard of. And it goes faster than any other tools they could use. So there we got a series of examples where yes, we can accelerate. It's not just about reducing your cost, but in some cases when your bandwidth is limited, is being able to go faster. And the the DR purposes, it's important. It means you can make your snapshot of the the uh replication much faster. So you if you had to recover from a uh disaster, you wouldn't have lost as much data as if you didn't use our product.

SPEAKER_00 16:04

Fantastic example of uh your solution in action. Um just let me ask you big picture. Uh, you know, we're we're being overwhelmed by AI agents now and new endpoints and sensors and connected devices. I just think about me personally. I've got you know an Aura ring, an Apple Watch, three eSIMs, Starlink, uh fixed wireless broadband, and as one person, right? I I could probably name 20 devices with data being generated. How far away are we from networks being just overwhelmed by this kind of IP connectivity and data uh uh creation? Is this a problem that we're gonna start seeing in actuality soon?

SPEAKER_01 16:50

It's a problem that you know everyone is aware of. Um and you know, I I remember in the in the days when Apple started to send data back from their phones and you know, and people didn't realize, and suddenly I couldn't watch Netflix anymore. And that's just because all our iPhones were utilizing the all the uplink bandwidth. And so when I was watching Netflix, my TV couldn't go back to Netflix and say send me the next packet because it was just congested. Um, so um what happens is a lot of people are aware of this. The the telco operators are uh very well aware, uh, and they they're asking people to be more you know uh frugal uh with the utilization of the bandwidth, but it's still a big problem. They on mobile devices like uh Aura rings or even phones, they are very also aware of limiting that throughput because it's a radio device and they're battery operated. So the more radio communication happens, the more the battery gets drained. So they are uh so the the the apples and the Android they are making sure that the applications are uh frugal on their communication. But this is where we actually uh can help as well, those applications to reduce the amount of transmission time and therefore help preserve the battery longer as well. So it's not just the networks being congested, but you have to think about the longevity of the battery as well. Um another example is uh uh recently I was discussing um uh with Oracle uh OCI uh here, the headquarters in Europe, uh, with DXC, which is used to the old electronic data systems, your own motors, etc., and we're discussing automotive. And in the uh some people don't realize the electric vehicles, um they have battery packs, and battery packs range from you know 400 cell to 4,000 cells in the Tesla. Um, each cell gets measured in terms of voltage, current, and temperature. So you may multiply that by 4,000 measurements, multiply that by lots of measurements being taken because hey, they need to make sure it doesn't catch fire. Um, and all that information is trended on the vehicle. And there is good research out there that shows if you can take that information and correlate it across a fleet of vehicles, you can further optimize the lifespan of your batteries, uh, remove you know uh events so the local battery management system, if it sees an overheating cell, it may shut it down, but it may be too late. Whereas if you correlate that information much earlier, you can manage the cell before it gets too late and you have to permanently shut it down. So there's lots of stuff that's um where you know, situations, electric vehicles is a very good example, and also in vehicles, all the assisted driving. So it's not just the only automated driving, but all these extra sensors and things, lane departure warnings, all of that. Um all that information can be collected, correlated to ensure that they are working optimally. That creates a lot of data, and which is most of the time still stranded on the vehicle.

SPEAKER_00 19:52

Yeah, fascinating. Every car is becoming a connected car now, not just EVs. So another challenge that you uh might be able to support, you know, as you talk to industry insiders and customers, partners, what what are some of the big misunderstandings or myths perhaps that folks have about Adam Beam and your approach uh that might need to be dispelled? I'm curious how what some of the reaction is. Change is always a little hard. It's a it's a it's a new approach, new architecture. Um how's that discussion going?

SPEAKER_01 20:26

I mean, so generally speaking, initially um the discussion is um it sounds exciting. That's the first reaction. Then the second reaction is sometime when you talk with uh IT teams and things, or well, you know, engineers, they say, well, we can do the same thing ourselves. We can ourselves. Um the reality is yes, you can, but not in the same way that we we we can do it. And so often the solution you you will put in place will be a lot more complex than just dropping our what we call the compaction tunnel transparent in line. You don't have to change anything, you don't have to change your applications and all that. And that that's whereas yes, you can do things yourself, but it's often it's sticky tape and wires, it's workaround, and the next time you have to do it again, then they do it differently. Um, our solution just we'll learn from the data. Some of the questions they go, well, what happens when the data changes? Well, when the data changes, we'll relearn, but how quickly do we relearn? Uh and so those things we in fact we've got new algorithms that will allow us to learn instantly as the data moves along and with very little uh processing requirements. And so these are are the things that uh no other systems can do. Um just and then the initially a lot so we said, oh, we transmit this code book. Well, how big is the codebook? And you know, so obviously the codebook, you know, you can make it as small as you want. Um, and in an IoT environment, like say the this trial that we did for the oil and gas, we created one code book and we then optimized it after 24 hours, and uh, and then after that it remained for nine months. Wow. Okay, so yeah, people think the data changes a lot. Actually, it it doesn't. Um, it really, but you have to in an IoT environment makes total sense. If you're in a telco router route routing data for millions of customers all over the place, that that wouldn't work there because it's all quite random and and so on. So that's why this this edge is is quite a good environment. A DR replication that works quite well because the workloads, the applications tend to be the same. Database replication, we also do 50% reduction. Uh so you know, all of this, if if you were to combine it across an entire telco network, it it you would have to do it in a different way.

SPEAKER_00 22:52

Interesting. Um well that was really insightful look into uh behind the curtain, as it were, into the technology at Adam Beam. What are you excited about over the next weeks, months, uh in terms of roadmap or uh deployments or just generally what is the team up to?

SPEAKER_01 23:10

Um yeah, I mean what I'm excited about is uh we're making we we're making some great improvement in terms of the uh maximum throughput that we can support. Um we're already at 250 megabits per second, and we know we're going to further exceed that. Uh, we've got further enhancements that will allow us to be also very efficient over uh high latency satellite networks and also further increase the throughput even uh for disaster recovery uh replication. So that that the R-site example I've given, uh, we will be accelerating that uh even further. So and you know, these sort of algorithmics around monitoring the data and adapting to the data in near real time, that's also will be coming up. So we the reality is there's plenty to be done uh with the product, and then once we've reached those improvements, then you know we will be looking at how do we shrink it and move it into microcontrollers uh and potentially eventually getting into um you know smaller devices. So we're gonna partnership with uh Trilliant, who are provide uh to smart grids and utilities, and they see our technology as being critical for what's called the AMI 2.0. So new metering, smart meters will collect a ton more data so that they'd be able to manage your smart grid, the electricity grid in a much more efficient way, especially because you now have solar panels, wind turbines, you know, batteries, all sorts of things. Um, that information is is very verbose, and they now see they partner with us because uh we're critical uh to enabling the collection of that information, taking it out of the smart meters. So we will incorporate our solution inside the communication module that gets attached to the smart meters. And that's our first step to integrating its small modules, and then we will continue that sort of low, low-level integration, which will eventually give us the ability to be, you know, our long-term view is we should be in pretty much any device and optimize the traffic, network traffic everywhere.

SPEAKER_00 25:21

Incredible. Well, that was quite a uh a unique opportunity there. Uh, thanks so much for sharing the vision and the mission and the opportunity. It really intriguing. Hope we can chat again in a few months and see how far you've progressed.

SPEAKER_01 25:35

Yeah, thank you, Evan. And and you know, great questions and uh yeah, real pleasure talking to you.

SPEAKER_00 25:41

Likewise. And thanks everyone for listening and watching. Also check out our TV show, techimpact.tv on Bloomberg Television and Fox Business. And uh till then, take care.