Mining for IoT/IT Convergence Gold (or Titanium): Q&A with Gareth Noyes, Wind River
What’s the best approach to realizing the benefits the Internet of Things (IoT) and Information Technology (IT) convergence can bring to industrial automation?
Editor’s note: Poet Dylan Thomas told us, “Old age should burn and rave at close of day….” But who’s to say how well any of us will succeed in following his admonition to “not go gentle into that good night”? At least we have a better handle now on how control systems will fare—and they don’t even have to burn and rave to avoid “death by lack of IoT.” One of the benefits of the new software virtualization platform, Wind River Titanium Control, according to the company, will be a new lease on life for legacy control systems. EECatalog spoke with Gareth Noyes, the company’s Chief Strategy Officer. Edited excerpts of the interview follow:
Embedded Intel Solutions: How are industrial systems influencing the IoT?
Gareth Noyes, Wind River: If we look at the impact of the IoT and the deluge of data that is coming, we see that future data volume is going to come predominately from things and infrastructure and less from people.
So while we see a growth in data driven and consumed by people, this will be completely overwhelmed by data from infrastructure (Figure 1).
And the infrastructure segments that we see will drive all the data are industrial systems: medical systems; smart factories; aircraft/avionics and so on. These sectors have some distinct functional characteristics. For example, control systems care a lot about time from a jitter, determinism, and latency perspective, whereas traditional software systems don’t.
A significant number of these systems may be remote and not accessible. Yet they demand high reliability and availability. These systems may be in place for a long, long time, and so the characteristics needed differ markedly from what we see in the personal computer and enterprise computer space.
Embedded Intel Solutions: How are companies trying to turn this impact that the various industrial sectors have on the IoT to their best advantage?
Noyes, Wind River: A lot of companies are focusing on an approach of: “Let’s just connect the things to the cloud and then have a way of accessing that data.” We have a portfolio of products that enable that strategy—traditional runtime operating systems like VxWorks and Wind River Linux connected to Wind River Helix Device Cloud for device management.
And that is a perfectly fine approach, but it doesn’t fundamentally change the way the business is built. An alternative way to merge IoT and IT is to virtualize the control environment and run on a compute environment close to the edge. It’s that merging of control and business systems where Titanium Control comes into play.
Embedded Intel Solutions: What will implementation of this new way to bring the IoT and IT together mean for Wind River customers?
Noyes, Wind River: Customers—and by customers I mean not only OEMs, but also the end users deploying factories and refineries—care because their businesses have very high integration and maintenance costs. In refineries, for example, having triply redundant control systems creates the expense associated with having a great deal of wiring. That wiring is literally hand configured with a screwdriver and a physical wire, so it’s very costly to move or upgrade these systems. Those high costs tamp down innovation. And because they are deployed for a long time, these systems become obsolete over a period of a decade or sometimes more.
Customers find themselves running very antiquated equipment in these critical control systems. With a better cycle of innovation customers can adopt much more modern technologies as ecosystems that innovate around their processes. They gain the ability to have open systems that enable interoperability among vendors. This is something they don’t have today because they tend to be locked into single vendor systems.
Embedded Intel Solutions: Merging with the IoT also has to mean protection from the cybersecurity threats associated with the IoT.
Noyes, Wind River: Yes, companion to all this is the industry’s reliance on an air gap model—the fact that their systems aren’t connected—to make them secure, and yet we know that is not a sufficient approach for things like Stuxnet.
So, the industry is looking for much more rigorous end-to-end and modern security models, and that needs to be addressed with the future architecture as well.
Embedded Intel Solutions: What steps are companies who want to bring legacy systems into the IoT era taking?
Noyes, Wind River: We are not alone [with regard to the perspective described above]. ExxonMobil has been very vocal in articulating the rallying cry to the industry to transform the present automation sector. And they’ve been very public about an architectural vision they have (Figure 2) where they are consolidating their control system architecture based on the Purdue [ANSI/ISA-95] model, and flattening that somewhat, so that the architecture Figure 2 shows is an amalgamation of several control function layers. The desired characteristics are high availability, real time, and having a software-based platform, not a purely hardware-driven control system.
Titanium Control addresses specifically this architectural element that Exxon talks about. You see the same drive in adjacent industries, including discrete manufacturing, medical, and transportation. We think the virtualization of control systems is an emerging trend in this sector, and we have seen similar trends in other industries.
We saw that in the avionics sector, with the advent of Integrated Modular Avionics (IMA)—something that Wind River is very familiar with—and we have been driving the same transformation in the networking and telecom space with Network Function Virtualization, virtualizing bespoke telecom equipment onto open, virtualized platforms, so there is a sustained transition that we have seen come into this industry.
Embedded Intel Solutions: What is essential to an edge-critical infrastructure?
Noyes, Wind River: First, infrastructure and control systems need to be extremely reliable and have high system uptimes. High availability is key.
These systems must be fault tolerant and be able to detect faults and recover from faults. That is a basic characteristic that we address with Titanium Control.
Once you have virtualized one of these control systems, one of the challenges you also must address is: How do you manage and orchestrate these virtual workloads? We will no longer be able to go over to a machine and the physical controller that machine has strapped to its side. Instead, that control system is going to be a piece of software running on a server. So, having the tools and infrastructure to manage and orchestrate that is key, and that is another area we’re addressing with Titanium Control.
Underlying all this is the need for performance and scalability—the systems must be able to meet the control loop and deadlines and schedules. There has to be fulfilment; there has to be tuning for latency, it has to be tuned for overall system performance. And we want “up and addward” scalability.
We want to be able to add more compute capability as needed and similarly, we want to be able to take away compute capacity if necessary. The scalability found in a cloud computing architecture is what we are bringing to control systems.
Last, but not least, all of this has to be based on a robust and modern security architecture, and so we take advantage of hardware-based security capability, including Intel® Trusted Platform Module (TPM) and root of trust and more modern encryption and networking security as well as user access and authentication control. It’s that ability to map a policy of who can access which part of the system and the ability to log and track that. It’s an IT-like model, which just doesn’t exist with physical control systems today.
Embedded Intel Solutions: What are some of the details our readers should know about Titanium Control?
Noyes, Wind River: The platform is based on open source components. At its heart, it is a Linux-based platform. The virtualization layer is optimized to address the latency and timing characteristics to make them more suitable for real-time environments.
One of the things needed under the hood in a virtual system, as you move workloads around, is the ability to migrate the data to match the workload. So, we have a very performant and tuned virtual switch, which moves the data as you move the workload. In addition to the compute-centric part of the platform just described, its other pillars include the control and management infrastructure.
That infrastructure is based on OpenStack, which some of your readers may be familiar with from the enterprise space. We can move a workload from one server to another as it is running. We can do failover, so the system can seamlessly fail from one physical server to another physical server—they may be adjacent to one another or they may be geographically remote—those are the sort of capabilities we enable with the control side of things.
With the amount of data coming through the system we need a very robust storage infrastructure, and we take care of that as well.
The vision here is that our customers and their ecosystems would be able to innovate on top of this platform and create virtual workloads. It could be a virtual or soft Programmable Logic Controller (PLC); it could be an analytic application—whatever other system our customers wish to virtualize. We believe our consistent API will enable the type of software innovation that the industry needs.
Embedded Intel Solutions: What differences will the architecture make to end users in the industrial space?
Noyes, Wind River: A lot of control systems today are bound by cabling and I/O. Virtualizing the control workloads allows us to extend the control loop away from the physical device to a more central location— possibly on the factory floor—so we could envisage a rack of industrial-grade servers doing all the control functions for a factory. This lowers maintenance and upgrade costs because the control systems are centrally located and allows looking across multiple control loops to get much more context about the end-to-end-system. It’s possible to look across multiple systems and get a better aggregate view of data. And this data still maintains context, because we can assign the control loop to the physical node on the factory floor, for example.
A more IT-like workload enables these OT-like control functions; it’s also a perfect compute node to be able to run our analytics. And we know that the vast data volumes and the functional safety and other requirements mean that much of the data in these critical infrastructure systems resides close to the source of the data, making them more likely to be off site than cloud- resident. So, this gives us an edge- or fog-type analytical capability.
When we look ahead to the impact of artificial intelligence and machine learning, which will come to this industry, as it will to many others, machine learning requires an [aggregation] of as many use cases and edge cases as possible. Aggregating from multiple control systems is critical, and we believe Titanium Control could be a great way of interfacing your control systems to higher order learning/machine learning. This is a forward-looking architecture that can enable that. The current architectures would not enable that.
We don’t foresee that this is going to be a wholesale switch in architecture. For greenfield designs this is a nice, modern approach that our customers could embrace, but we also know that we must be compatible with or be able to work alongside legacy systems, and Titanium Control allows us to do that. One use case might be enabling plant-wide system simulation with our customers—enabling them to have a twin operating in real time in conjunction with a physical asset, and having a much easier migration path from the simulated world to the real-world control function.
Embedded Intel Solutions: What will the impact on OPEX and CAPEX be?
Noyes, Wind River: We believe we can drive down the operating costs of these types of control systems, whether through reduced installation and integration and commissioning costs or via lower maintenance costs. You can monitor these systems remotely; they are more centralized; there are a lot of factors that we believe drive down the operational costs. The Titanium Control software virtualization platform is based on a standard computing hardware and software model, so the upfront CAPEX costs are going to drive down as well.
This leads to a shorter time to market for customers in the control space. It also drives down the complexity of the systems, giving them insight as to how the overall control system is working.
And, having de-coupled the software from the hardware, it is going to be much easier over time to take advantage of new silicon processes and software advances and be able to upgrade the system as technology develops.
Embedded Intel Solutions: Does the platform leverage software’s ability to now take advantage of some Intel hardware features that may not have been fully capitalized upon earlier?
Noyes, Wind River: Being able to take advantage of root of trust with the TPM, encryption acceleration, things like that—we leverage those with this platform. There is also the concept of enhanced platform awareness, where we can take advantage of capabilities in the silicon such as the libraries for packet acceleration and being able to do higher availability low latency networking—system manageability security features that are enabled through this enhanced platform awareness interface. Hooks in the software allow it to take advantage of underlying features [in the] hardware, and we have taken advantage of many of them—manageability, power management, performance, and network switching—to name some.
Embedded Intel Solutions: How might the features of this virtualization platform be introduced to a factory, for example, in a way that allows incremental implementation?
Noyes, Wind River: One way is to have the virtual platform be a failover for a physical platform—if you can connect to the same data bus, then we can either mirror or take over control functions entirely. The sort of things that we are also looking to drive, though, are the orchestration and manageability of actual physical workloads, so we can start doing discovery of physical infrastructure into our virtual orchestration as well. Those are the sort of capabilities we are looking to drive into the platform that will require interoperability between Titanium Control and operating systems like VxWorks and Wind River Linux. We are looking to integrate those types of features into our portfolio.
Embedded Intel Solutions: You already have design wins with the Titanium Control virtualization platform. Please speak about why those wins happened and as well add any closing thoughts.
Noyes, Wind River: We see interest from continuous processes and process automation. Our customers believe they can get product to market quicker with lower overall system and maintenance costs. The secondary driver is customers gain the ability to build innovation on the platform and drive new revenue sources by offering a new service model and creating new analytics and predictive maintenance models.
We’ve articulated a vision that has resonated very well with the industry, and having champions such as ExxonMobil articulate the same vision has been very helpful.
We have seen a great amount of interest in particular from the process automation and control sector. An emerging group, Open Process Automation Forum [part of Open Group] is aiming to drive standards around this, so there is already an industry-wide consortium of end users and suppliers, of which Wind River and Intel are a part.