Edge cases
Dell Technologies’ Jason Shepherd joins Jonni Bidwell in a stylish-looking stairwell in Edinburgh to talk about a de facto standard for edge computing.
Dell Technologies’ Jason Shepherd joins Jonni Bidwell in a stylish-looking stairwell in Edinburgh to talk about a de facto standard for edge computing.
Jason Shepherd is a Governing Board member on the Linux Foundation’s LF Edge project, as well as CTO for IOT and Edge Computing at Dell Technologies. Back in 2018, at the Linux Foundation’s Open Source Summit he was kind enough to spend some time to talk to Jonni Bidwell about the Edgex Foundry project, a common open platform for computing at the edge of the IOT.
Mediating between the cloud and IOT devices is easy, but devices may lock you into proprietary edge devices or their own private clouds, and public clouds lock you in to their APIS. The challenge is to come up with open APIS which enable different hardware to interoperate and to get manufacturers to agree to use them. Edgex Foundry has, in its short life, achieved precisely this. As it happens, Edgex has recently celebrated its 1.0 release, which shares its codename with Scotland’s chilly capital.
That great city where once a young Jonni Bidwell was often to be found in a pub in Portobello, asking his beer and/or whisky what one is to do with an undergraduate degree in mathematics. On this occasion he and other distinguished open source luminaries enjoyed a whisky reception at Edinburgh Castle, proving conclusively that there is life after maths.
Linux Format: Hi Jason. I’m Jonni, technical editor for Linux Format – an actual paper magazine, would you believe.
Jason Shepherd: Hi Jonni. I’m Jason, CTO for IOT and Edge Computing at Dell Technologies. My role with my team is to drive our core strategy, so we look at market strategy for IOT and Edge. We own this engineered roadmap where Dell Technologies is building infrastructure, from the thin compute edge, through the mini edges to the mini clouds.
One of the problems right now, and we’ll talk about this more in a moment, is that everyone is combining applications with the core infrastructure, and that’s not going to scale. When’s the last time your ERP system managed your PCS? Bad idea – you’ve got to decouple this stuff. My team is about “How do we get to the scale factor? How do we drive core strategy?”
And of course standards in open source are a key part of that. I got Edgex started with a phone call in July 2015. I called a team at Dell and said “This isn’t going to work”. I was driving somewhere between Santa Clara and San Leandro doing the magical mystery tour, meeting all these companies trying to figure out how we were going to scale this thing. I realised we needed the right architecture, so I called up my team and said “What if we tried this?”
Admittedly, we initially considered creating a Dell platform, but nobody wins if we’re the 401st IOT platform out there. Plus it’s not just about basic interoperability, it’s also about what I call the holy grail. And that’s the holy grail of digital, not just IOT. A lot of people are just getting going in this market, but IOT has been done for a long time in embedded systems – it’s a buzzword now. I mean, how often do we say “I’m going to do some e-commerce today?” No, you just buy stuff online.
LXF: Our marketing department have much to learn, apparently.
JS: IOT, or any of this edge stuff – they’re all buzzwords, but it’s about solving customer/end user problems, getting better experiences, using your powers for good not evil, and solving real-world problems. It’s been going on for a while, but right now it’s all about hitting scale.
The holy grail is selling data – everyone says data is the new oil; resources – compute, storage, networking, energy, ride-shares… y’know, anything consumable; and services – your domain knowledge – to total strangers. So take an open source expert, why can’t they go and consult for someone who’s never met them before? I’ve had various conversations with many smart people and no one’s said that’s what they ultimately want. They want to share or
monetise data resources and services with complete strangers. You will never, ever get there with a bunch of lock-in models, you must have openness.
In consumer, trust is built, generally, with a number of entities that you get value from. I know my UPS guy by name. Amazon’s great. Four years ago I told my team “Hey, Amazon’s gonna win the first round of consumer – don’t even try”, because they’ve got the double whammy: they sell you content and stuff. Google will do fine, Apple will do fine, and then you’ve got your Facebooks and such. Sometimes the trust is violated, especially around privacy, but generally, in consumer, if you build trust with someone and you get value from their services, then your privacy goes out the window. Period.
In business, or in business-to-businessto-consumer, or crossing between any public and private domain, you cannot have any single owner of the trust – it doesn’t work. So the current IOT market has everyone building these platforms that lock people into their clouds or whatever, so they can make money off those people’s data. That’s completely backwards. Since all data is collected at the edge, what you need to do is set the data free the moment it’s created, and use open technology to bring back checks from strangers. This is partly why open source matters so much. You know the movie My Big Fat Greek Wedding with the guy that sprays Windex on everything?
LXF: Of course, it’s a classic. Windex cures ills from psoriasis to poison ivy.
JS: Well, I see blockchain as one of the Windexes of technology. And 5G is another one. These are important technologies, but they won’t solve the world. This will take 3-5 years, but there are already companies like Helium, who provide shareable Wi-fi bandwidth. That’s the resources part of it. A lot of people just talk about the data, but it’s really about data, resources and services.
It’s about selling those to strangers or, in a more altruistic context, sharing them. You can’t get there without a combination of technologies. So what you need is industry collaboration, open collaboration. You need root of trust for the silicon level. All data in a digital sense is all created in silicon: things or devices or PCS or whatever. You need trusted zero-touch provisioning, and by that I mean when I ship a box somewhere, I don’t know who’s going to plug it in. I still need it to get the certificate and the dial tones securely and safely. You might’ve noticed that Intel and Arm announced that they’re collaborating on this. You need that industry collaboration.
Then you need open frameworks like Edgex that are architected to bring together an ecosystem around very heterogeneous components, because that’s what embedded is. We’re collaborating in this new Akraino project – that’s a telco-centred infrastructure project. Edgex is at the application layer, Akraino is more infrastructure-level. So great things happen when you collaborate there. Then, yes, you need some Ledger, and maybe some AI for contents. And when you combine all this in the right way, with industry collaboration, no single ownership of trust, then I can literally create data in the physical world – and if I choose to, then I can share it or sell it to complete strangers. This is scale.
I’ve never met anyone who has said this can be done the way people are doing things today. It won’t work, and people say “Oh, that’s logical”. In the ocean you have riptides, and people’s natural inclination is to swim against the current back to shore. If you try this you’ll get tired and you’ll drown. In any new emerging market you get this herd mentality, and just now that’s all about lock-in models.
What we’re doing as an open community is we’re swimming sideways, which is what you’re supposed to do. You collaborate, you get out of that lock-in, and you start to open things up. In a similar water analogy, I would say the maker movement is why IOT came along. It used to be with industrial, buildings, manufacturing, whatever, they’d be proprietary platforms that would lock you in and cost you tens of thousands of dollars a year.
The moment an innovative small company pops up that threatens your way of life – and this has been going on for years – [they] buy ’em up and kill ’em. But with all of the power from this open collaboration, and the kickstarters, there’s not enough money in the world to buy up all those innovators.
You either change or you die, and this is the innovators’ dilemma. So we’re all about floating all the boats for open scale and collaboration and, in this world, just make sure your boat is really good and really fast. You’ve got to innovate and stay ahead. From Dell Technologies’ standpoint we gave away our code to seed the project. The Linux Foundation has to be completely transparent – they do a great job running transparent, open projects with governance when needed.
LXF: They do a great job organising these interviews for me too. There are an awful lot of distractions, both of the technical and sugary nature.
JS: We knew that this needs to be open. I’d say we’re at the ‘AOL’ stage of IOT: it’s early days. Initially Edgex is about interoperability. The easiest way to describe it to people is it will do for IOT what Android did for mobile. You have to abstract that interoperability of the
function out of the OS, out of the protocols, because there will never be one protocol. You need a de facto standard and de facto standards are created through consensus – and a great way to get consensus tangibly is with open source code. You need to make it Osagnostic, hardware-agnostic, protocolagnostic, agnostic to Kubernetes, Mesos, Docker… We don’t make choices on infrastructure in Edgex.
LXF: I feel it’s hard for a lot of hobbyists or app developers getting into IOT to see why a middle layer is necessary. They have a small device like a Raspberry Pi that we love so dearly, and that can juggle data locally fine – and if the application needs some kind of centralised data management, there’s varying degrees of services and clouds that are available commercially.
JS: I think for lots of the new IOT use cases people are in what I call ‘Pi in the sky’ mode. I take something like a Raspberry Pi and I connect it to a public cloud, and I don’t know yet that I’m going to have a problem with pumping data mindlessly into the cloud, or that I’m going to have to pay egress charges to get my own data back. So you want to decouple the edge from the cloud quickly. The clouds – and some of them are doing great things, I don’t want to be too negative about them – but they’re trying to give people IOT gateway drugs. Y’know: let me get you hooked on my cloud… here, have a dev kit that makes it all too easy to get locked in to my APIS. Edgex Devkits will give you choice as a developer to work with other cool projects like Akraino, or things like Blockchain and Ledger.
LXF: So in a lot of ways it’s quite abstract and imaginary, or up in the air, in the sense that it doesn’t care about what it’s running on? Traditional embedded computing is so tied to a particular bit of hardware, for reasons of smallness. I’ve been hearing ‘cloud native’ a lot lately. JS: Right. So Edgex is a cloud-native framework, pushed down or shrunken so it can run on something like a Raspberry Pi or Beaglebone or whatever. The thin compute edge, I would call it. There will still always be embedded computing, but the thin compute edge is basically the last point before you get into controllers and other embedded stuff like sensors, very constrained microcontrollers and PLCS in the manufacturing line. It’s the last point where you’re still software defining workloads through containers. It’s driven by memory – it takes about 256MB where I can run a hypervisor and Edgex, or something like it. Maybe some security and management plug-ins, maybe a basic rules engine, but you’re not going to be running massive deep learning on this kind of hardware.
We’re around that level with the footprint today. When we launched, our prototype code was about 2.5GB – it was huge. But we had to tell people not to worry, everyone agreed on the architecture: loosely coupled microservices, polyglot. So we took those principles, and made it optimised for nodes, things that don’t have a person in front of them that are also distributed all over the place. Security, management… it’s protocol soup down there, where as up in the IT world there’s only tens of protocols that matter.
In the community we switched from the Java prototype code to Go. Obviously Go’s a very popular language, it allows you to maintain platform independence. I mean
IT’S A LOSING BATTLE “Generally, in consumer, if you build trust with someone and you get value from their services, then your privacy goes out the window. Period.”
you could make it tiny in C, but you’d lose that independence and it won’t scale as much for this ecosystem. But it’s polyglot too: we have our reference implementation in Go, but you can replace anything in there with whatever language you want. It’s bound together with the APIS that are governed by the technical steering committee and the Linux Foundation. Driven to be a de facto standard by new people piling in.
Edgex started with about 50 founding companies. We built that up over a number of years, building consensus with a lot of great folks. Our codebase went down from 2.5GB down to 128MB in about 18 months, which is pretty good (can they do the same thing with the Ubuntu ISO,
please? – Ed). The beauty of a microservices framework is that individual people can work on different functions independently and then bring them back together again.
We knew from Dell, when we got this going three years ago, we knew it would take about three years for the big companies to get on board. I figured they’d be trying to own everything for about three years, for all the aforementioned reasons. But now those companies are realising, and not only just for the basics, but also the holy grail, that they need to collaborate on just enough plumbing to interoperate. And not only just to get going and start building things. Do you think the PC market would have scaled if it cost $1,000 to connect your keyboard? What if every website required a custom protocol?
No, you make your money on the idioms: security, manageability, scalability, connectivity, usability, augmented reality – you do not make your money on the plumbing. And this is why open source – and I guess I’m preaching to the choir here – helps you stop undifferentiated heavy lifting.