Zededa creates a new architecture for the edge

Edge computing efforts are a dime a dozen nowadays, but after mocking the launch of Zededa a few weeks back for its buzzword-heavy press release without any technical details, I talked to the company’s CEO, Said Ouissal. He explained exactly what the startup’s vision is, and it’s a pretty novel way to build applications that could run on any gateway device.

The overarching goal of the startup is to help companies put software on edge devices that can be run securely, with little expertise needed from customers. Those devices might be machines aggregating sensor data or traffic cameras monitoring a street. Ouissal sees each of these devices as a set of infrastructure with common traits, which means Zededa developers can build applications that can span many different devices — and needs.

Zededa’s approach is akin to how developing software for the cloud works today, except that with the cloud the underlying physical hardware is relatively similar. In the IoT world, there are seemingly endless different types of computing devices — from a $6 Pi-based computer to a powerful Xeon gateway. There’s also a big question as to whether one needs to build vertical solutions for the industrial world.

A large contingent of industrial IoT entrepreneurs are betting that customers want to buy hardware, software, and cloud services that are vertically integrated so they don’t have to manage complex IT sourcing for something that could become vital to their business. Zededa thinks heterogeneous hardware and existing customer hardware can instead be transformed into something that handles a wide variety of applications. It basically wants to become the Amazon Web Services built on top of millions of connected IoT devices.

To do this, Zededa is creating a software package that combines a hypervisor and a new concept in computing called unikernels. Unikernels are packages of software that contain an application and only the underlying operating system required to run that application. So if the application doesn’t need a file system, that gets jettisoned. The end result is very simple blobs of code (I’d call it a container, but that means something different).

A container, such as those offered by Docker or Kubernetes, provide everything a piece of software needs to run such as the OS, runtime, libraries etc. It’s more flexible than a virtual machine created by a hypervisor, but has more overheard than a unikernal.

The hypervisor is important as well. While newer IoT implementations might view hypervisors as a relic of the server era, there are millions of older connected computers running Windows operating systems that can’t be shoved in a container. For those, you need a hypervisor, says Ouissal. He’s not alone. Last week, The Linux Foundation released an open-source hypervisor designed for the IoT with contributions from Intel and others. It’s called ACRN.

These elements communicate exchange data with the machines they are on and also send information back to a cloud operated by Zededa. The blobs of code and the hypervisor help ensure that the applications that are accessing the edge device stay secure even if the device is tampered with, while the cloud governs the way applications run on the extended hardware devices.

Some of this approach reminds me of what Resin.io is doing with its ability to run containers at the edge, allowing customers to manage applications across their fleets of IoT devices in a way that’s closer to the way they can manage their applications across a cloud infrastructure. But a lot of this also feels very novel, such as the adoption of unikernels that allow software to run in constrained environments.

I’ve spent years trying to define an edge computing stack, and it shifts depending on who I talk to. The one constant, though, is that it’s trying to use existing technology to solve what feels like a very new computing paradigm. And I’m not using the word “paradigm” as jargon. Creating a trusted, secure, auditable, and manageable way to deploy software across millions of nodes is a very different challenge for computing. It really is a new paradigm.

I’m not sure if Zededa’s software is the right path forward, but when it launches later this year, I can’t wait to see how people build on it and with it.