Edge computing powers devices to process before transmission
Mike Bell is the executive vice president of IoT & Devices at Canonical. He joined the company in October 2016 from car maker Jaguar Land Rover’s connected car and in-car entertainment business. This background gives him great familiarity with embedded devices as Canonical focuses its IoT and devices efforts on the mobility, industrial, networking and digital signage sectors with its Ubuntu open source software platform. Here, he tells George Malim that more and more intelligence needs to be available in device locations and edge computing is enabling that in a secure, cost-effective way.
George Malim: How does Canonical see edge computing developing in terms of how it is enabling Internet of Things applications and services?
Mike Bell: The development of edge computing is highly positive. Ultimately, there’s a huge amount of hype in the whole IoT area but the edge device or IoT gateways are specifically the area where I think we’ll see a huge amount of growth. If you take the oil and gas or industrial manufacturing sectors as examples, all have their own operational technology (OT) networks but, as cloud becomes beneficial to those users, the internet is punching holes in that OT architecture. Organisations therefore want a bridge between the OT networks and the internet.
However, that doesn’t necessarily mean that everything gets sent to the cloud for processing and storage. There’s a need to do more processing locally, partly because of latency. In addition, a lot of the data is only useful for a few seconds so it needs to be processed locally to maintain its value. The market gets this, we’re seeing proofs of concepts where people are looking at locally-based machine learning.
Cloud is still, of course, a big part of the IoT equation but we’re seeing more compute power being pushed out to the edge and that can be server-sized in terms of the CPU capability being included in edge devices.
Another area that involves a similar class of device is telematics. A telematics unit is generally an aftermarket device, deployed with cellular connectivity to enable monitoring of a refrigerated commercial vehicle, for example. We’re seeing more use of aftermarket devices like this and edge computing as a general concept can be used quite broadly because it offers separation from the edge of a [communications] network and the cloud or an enterprise IT network.
GM: Edge computing seems to be an inversion of the cloud pitch of removing intelligence from remote locations and centralising it in the cloud thereby enabling unintelligent commodity devices to be deployed. How easy is this to communicate to a market that has been fed cloud messages for many years?
MB: Edge computing is about applying intelligence locally and that’s vital to the success of many IoT apps. If I look at the data points coming off a car, there are gigabytes of data flowing around at any point but there’s no ability to move that off the vehicle. The vehicle therefore needs massive storage capability in order to hold information and wait to stream it later by pushing it to the cloud. The car has a lot of data but it’s too much to move over a cellular data connection.
Edge computing can be used to apply intelligence to the data and isolate valuable or urgent information and then send it to the cloud via the cellular network. You could argue that edge computing is now relevant because of the availability of network access and the cost of wireless infrastructure. And it’s not just applicable to devices that change location. For example, if a water company was monitoring a lake, a device could be fixed on the boundary between two cell sites with the signal dropping and switching between the two so continuous connection isn’t viable. Edge computing would enable that device to have a level of storage so it can send data when connectivity is available.
GM: Monetisation is one of the greatest challenges facing organisations that are deploying IoT services and many IoT apps still appear to be experimental shots in the dark. How do you see organisations monetising IoT effectively?
MB: At the stage we’re in, I think progress can be judged based on what device makers are doing. If they’re selling more boxes, the market is picking up and greater volumes mean people are moving from design and proofs of concepts to larger, more commercial deployments.
We’re a horizontal platform play, we want to allow an independent software vendor (ISV) to separate its intellectual property from the hardware it sits on. Hardware is a race to the bottom on cost and margin and the value is in the services and software. Companies today therefore are selling solutions as a combination of hardware and software products. This is being done as a means to an end to sell customers a box with software, they don’t really make money on selling the box.
These companies like our concept of snaps – a universal Linux package format that works on any distribution or device – to disaggregate hardware from software. We’ve done this with a switch manufacturer which has taken a whitebox switch and applied their intellectual property – in this case their routing capabilities – as a snap. By doing so, they’ve effectively disaggregated the switch from being an appliance to being like a server.
We can see places in IoT where this is happening. A lot of customers like the idea but don’t generally launch with it, largely because they already have hardware and software solutions in place. However, we’re providing a supported operating system for IT and that’s the missing piece of the jigsaw.
We see a value play in providing a stack that builds up with different snaps in our store and people are starting to embrace that. We don’t want this to become like an Android or Apple store because each device is so different but we want customers to create their own devicespecific stores. We do want to restrict where software can be downloaded from for security reasons.
GM: How challenging is it for organisations to construct a business case for IoT when monetisation can be hard to establish? How, for example, does a traditional products company track the business case for servitisation?
MB: Very few people are moving from selling products to selling services now but we definitely see this becoming the norm. Enterprise customers want to do edge analytics and machine learning and bring in customer feeds but these are the same as a traditional enterprise IT business case. They’re no different, a business case is a business case and organisations have to work it out.
For companies looking to move from providing products to service cases, this can be more challenging. For some, the business is the product while others see opportunities to move on by putting intelligence into the product. That could be connectivity or a screen but it has to add value to the product.
For example, it would only be worth adding this intelligence to a high-end fridge because people are willing to pay more for a feature. Having worked in the car industry, I’ve got a strong understanding of how need and want are very different things when it comes to what people will pay for a feature but it’s clear some customers will pay a lot for features that they want.
GM: IoT is unlike other IT areas when it comes to security because there’s no easy-to-identify perimeter and the threat surface is massive with inherent weaknesses in the multiple handoffs between devices, networks, cloud and servers. How should IoT security be approached, is it really any different to IT security?
MB: I think the fundamentals remain the same. The threat surface is the interesting point because a server in a data centre can have firewalls, anti-intrusion detection and physical security but something on a wall of a building will probably not for reasons such as cost. The security technologies for IT and IoT are the same but the difference is that people have deployed IoT devices without installing security or because the cost of securing devices could make the product commercially unviable.
There isn’t a specific way to secure IoT. All the same security you would apply to IT is appropriate but there are more threats in IoT. Even so, no new technology will solve it, classic architectures like a defence in depth strategy will be the best approaches. You can build and architect for security, though. If you choose Ubuntu Core as an operating system, we have built-in security from the ground up.
GM: Do you see security capability as a means for IoT technology vendors to differentiate themselves?
MB: Yes, it’s definitely a differentiator. We recognised that if you have a smart Wi-Fi camera on a building, the idea of sending someone with a ladder to update it with a USB stick is ridiculous. The technology we have for updating server technology is exactly the same for IoT. Our approach is to constrain the kernel and enable the stack to be kept up to date over the air.
The other point with over the air updates is that if something goes wrong such as the operating system breaking a device, Ubuntu Core would recognise it cannot boot and go back to the last known image of the device. This means the service would run while a fix is applied via transactional roll-back, over the air update and the containment of the device.
GM: How do you see Ubuntu developing in IoT?
MB: Ubuntu as a whole is getting very interesting and we’re seeing different classes of deployment from laptops and PCs to the data centre. Companies are asking themselves why should they maintain a custom version of Linux for their appliances and others, such as PC manufacturers are paying us to enable more and more of their hardware.
The idea of there being a Linux desktop for every person in the world never came off but world leading innovators are using Ubuntu as their developer desktop. Whether in robotics or machine learning, Ubuntu is being utilised and in self-driving cars NVIDIA’ s offering is based on Ubuntu. Very smart technology and device companies are adopting the technology so we see real momentum for Ubuntu across all classes of devices.