Typhoid Mary & the Spread of Real-Time “Deterministic” Networking

Authors: Michael Levy and Glen Allmendinger

Until quite recently, “real-time” or time-sensitive networking existed only in unique domains where applications demanded extremely precise timing and synchronization: Robotic manufacturing, machine vision, automotive infotainment, smart transportation, oil drilling, space exploration, and the like. The protocols were proprietary and technically difficult, and were often fine-tuned specifically to mission and safety-critical applications and industrial machines.

Now, with the rapid development of the IIoT, these precision networking capabilities will escape their “walled-garden” confines to operate in society at large, where they will enable autonomous vehicles, real-time road sensing, high-velocity supply chains, personalized biometric banking, and many more applications that most people are overlooking today. This is because real-time networking will spread much farther through society, and much faster, than most people imagine.

A valid analogy might be to the patterns of contagion, which first became well-known in the late 19th century when an Irish cook named Mary Mallon (“Typhoid Mary”) was identified as a carrier of typhoid fever who had infected numerous people. In a similar way, adoption of real-time networks will spread quickly among manufacturers of networked machines, equipment and devices, whose innovations will geometrically spread into diverse and unforeseen application-areas. Those who are able to leverage the economies of scale and componentry, as well as the necessary skills across domains, will become very successful indeed.

WHY WE NEED REAL-TIME, DETERMINISTIC NETWORKING

Real-time or time-sensitive networking has existed at least since the 1970s, but until now it has been confined to the “walled gardens” of extremely esoteric, safety-critical domains: Robotic manufacturing, machine vision, automotive infotainment, space exploration, and the like. Real-time networking is important because in many human endeavors, a sluggish network connection is not merely an inconvenience. If a human worker’s hand suddenly enters the path of a gigantic robotic arm on an assembly line, the robot has to stop instantly—in “real-time”—to avoid harming the worker.

But what exactly does “real-time” actually mean? Does the robotic arm have to stop within 30 milliseconds of sensing the worker’s hand? 10 milliseconds? 10 microseconds? Engineers will tell you that the requirement is always “whatever your specific application needs it to be.” The robotic arm on an assembly line may have the latitude to stop considerably slower—measured in fractions of a second—than a robot used in remote open-heart surgery. In any event, there is always a drive to approach zero latency, which explains the recent expansion of the Ethernet standard to include time-sensitive networking (TSN).

Companies, researchers, and thought leaders are now zealously predicting powerful applications and use-cases, starting with industrial automation and autonomous vehicles and going to the moon from there. However, these visions demand a technology that’s much more agile and precise than anything we’ve seen before.

In the walled gardens of the past, proprietary solutions from players like Siemens and many of their industrial peers allowed signals to travel from the physical connection (usually Ethernet) up the stack to actual applications. Those protocols have worked well enough in the tightly controlled environments of the walled gardens. But they will not meet the requirements of networking in the open world, where real-time connectivity will be central to any number of innovations, including AI at the edge—which will itself control surgical as well as manufacturing robots, and a great deal else besides.

A BIGGER OPPORTUNITY THAN MOST PEOPLE THINK

In the open world-at-large, the keys to the kingdom will be true interoperability, and how quickly it can be achieved. If only for that reason, having a single open standard is greatly preferable to a multiplicity of standards.

To date there have been three traditional factions involved in networking:

• The IT crowd, which operates on a legacy “batched” information processing and network architecture that was never real-time;
• The telco and wireless carriers, who have built the largest real-time networks in the history of the world. These carriers know how to connect only two things to their networks: cell phones and the modern-day computers we call smartphones. But they have not connected those phones in a time-sensitive or “deterministic” way. Mobile phones always have a human being attached to them who can respond gracefully (i.e., in a human way) when their device stops working on a subway. Unmanned sensors cannot do the same, and most telco people wouldn’t know a sensor if it hit them over the head;
• Automation and control players with their walled-garden history.

Today, these factions are all, to varying degrees, living in the past. Their current networking technologies will not support open and interoperable real-time networks, and the emerging world is virtually screaming at them, “Stop dragging your historical box of rocks around and get with the program!”

In 2012, the concept of time-sensitive networking (TSN) emerged as a deterministic Ethernet-based network standard which the world has come to see as the “silver bullet” for industrial IoT. In response, industry players have poured millions into developing the end-to-end real-time solution of the future. Since TSN was introduced, many sub-groups within the overarching standards committee have been examining applications in many more places than just the industrial arena. In fact, the non-industrial applications world is now driving the expansion of the standard to cover more and more real-world possibilities.

FUTURE SUCCESS = ONE STANDARD FOR EVERYTHING

TSN is a set of official standards established under the umbrella of the 802.1 working group in the IEEE 802 standards committee, but it exists only on the data-link layer (layer 2) of the 7-layer OSI network model. On its own, TSN can’t go below its own layer to the “field level” of physical sensors, or above its layer to the level of applications. It needs to be paired with a protocol that can communicate with other devices as well as with the gateway.

This is why industrial OEMs with existing investments in their own legacy protocols (e.g., ProfiNET, EtherCAT, EtherNet/IP, and the list goes on and on) have poured huge resources into positioning their protocols as the key to unlocking TSN’s capabilities, thus creating an arms-race between OEMs and industry organizations.

While TSN supports real-time control and synchronization over one Ethernet network, it can also support other common traffic found in manufacturing applications. This capability is driving convergence between IT and operational technologies, which in turn sets the stage for TSN’s expansion beyond just industrial applications. TSN is not a disruptive technology. It enables existing industrial real-time control applications to coexist and interoperate, such as the new OPC UA publisher/subscriber model currently being finalized in the OPC Foundation.

The proper way to think about real-time networking is that we do not want a “multiplicity of standards to choose from.” We want one standard that works for networking everything. People who understand this, and who give up the protocols of the past and pay attention to the great need for interoperability across the spectrum of domains, are likely to make a great deal of money.

AUTOMOTIVE BECOMES THE FIRST REAL NON-INDUSTRIAL BEACHHEAD

The automotive market differs from industrial automation in that one protocol, CAN, has significantly more real-time capabilities than other protocols like LIN, MOST, and FlexRay. In addition, CAN’s low cost of installation is very important in the extremely cost-sensitive automotive market. Only luxury brands are able to implement the more-expensive FlexRay.

New automotive applications, especially those designated “safety-critical,” have extremely low latency requirements that CAN and not even FlexRay can meet. CAN maxes-out at 20ms, whereas TSN, at its worst, starts at latencies of 1ms. In addition, in-vehicle infotainment is a major use-case for TSN. TSN’s predecessor protocol, AVB, is already implemented in cars for infotainment, but TSN has the ability to prioritize network traffic so that infotainment can coexist on a single network with applications like ADAS (Advanced Driver Assistance Systems) without interfering with them.

The introduction of TSN will not only change the automotive industry, it will drastically change the physical architecture of the car. The adoption of Automotive Ethernet and TSN will necessitate a “brain,” or data aggregator, connected to each application. TSN will first be used for safety-critical applications and infotainment. As hardware costs decrease, it will completely replace CAN in vehicles.

One major barrier to adoption for TSN in the automotive space is its reliance on Ethernet. For TSN to be widely adopted, Automotive Ethernet has to become pervasive in the industry. But Automotive Ethernet is currently expensive, meaning that TSN will first be adopted by luxury brands (i.e., Audi, Tesla, BMW) that can absorb those costs. Despite this fact, the potential of TSN to unlock real-time automotive applications has spurred the industry to rapidly shift towards Automotive Ethernet—representing a vast opportunity for suppliers and developers alike. One obvious aspect of this shift is that it will drive componentry costs down. Volvo, a key proponent of Automotive Ethernet, is especially well-positioned to succeed. As we move into the next decade, Harbor Research estimates that Automotive Ethernet alone could enable hundreds of millions Ethernet ports on vehicles.

“TYPHOID MARY” GEOMETRICAL GROWTH WILL DRIVE REAL-TIME NETWORKING DIFFUSION

As TSN and real time network technology dovetails with other adjacent developments in general networking, it will help enable and accelerate ongoing efforts to bring higher bandwidths to more diverse devices. More importantly, the push to automate network configuration and management—often referred to as Software-Defined Networks (SDN)—and SDN’s ability to enable automated programmatic configuration and management of networks, will remove much of the complexity from using these networks in the industrial ecosystem, as well as in mission-critical commercial applications like autonomous robots and lift vehicles in warehouses within larger supply chains. As real-time networking spreads to non-industrial applications, deploying these capabilities will require, more than anything else, simplicity.

How far and how fast will these real time networks spread? Will wireless become more deterministic over time, and what are the potential impacts of 5G on real time networking? How should developers of networking technologies think about wireless and real time networking?

To answer those questions, consider the networking that you and everybody else uses every day: wireless local area networking, or Wi-Fi. Wi-Fi “works fine” for your laptop and phone while you’re in your home or office, and if it’s configured correctly it’s even reasonably secure. But if you think that Wi-Fi is “good, reliable networking,” you have another thing coming. Wi-Fi isn’t even slightly “deterministic”—i.e., there is no guarantee of performance at any level.

Similarly, if you’re one of those people who thinks that 5G is going to solve world hunger, you really have another thing coming.

The technologies behind 5G enable the use of very high frequencies. The higher the frequency, the shorter the wavelength. Shorter wavelengths enable faster speeds and lower latency. But that’s not the whole story. With shorter wavelengths, the distance between the device and the “tower” has to be shorter, and the signal itself can’t penetrate dense materials like walls and trees. To get around these deployment challenges, providers will need to deploy vastly more towers.

In order to have reasonable coverage, providers must also build 5G antennas and towers everywhere, and very close to users. This is a time-consuming and expensive process that will make its rollout slow and uneven. When carriers say they’re rolling out 5G in a city, what they mean is that 5G will be available in limited pockets of that city. The consistent, fast, and reliable 5G dream that everybody talks about will be available in some offices, entertainment venues and other locations, but not generally.

So, how will 5G relate to TSN and real time networking? Engineers in the telco equipment development arena are speculating about using TSN to handle the distribution of network traffic between 5G cell towers to ensure real time performance and no packet loss. But the reality of this story is that Rome wasn’t built in a day; five years from now, your smartphone will use 4G most of the time, even when you’ve got a 5G phone in a 5G city. The wireless carriers hope that 5G will enable them to compete with, if not replace, ISPs, cable companies, and satellite internet and TV companies. In actual fact, it will probably be more than 10 years before 5G replaces 4G for most users most of the time.

Where is real time and mission-critical network heading? We believe it’s heading to many more places than most people imagine. Of course, real-time deterministic networks will address safety-critical applications before they spread to mission-critical applications. But think about all the real-time problems that occur around us all day—getting accident victims to hospitals, processing multiple data feeds, enabling advertising and wayfinding capabilities for individual consumers, tailoring advertising on kiosks in brick and mortar retail locations, and many more applications to be sure.

Consider “multi-modal” sensing, which combines multiple types of sensors and data feeds that can be fused in real time. A simple application example could be the combination of video and auditory sensing for safety and security in a smart city setting where, if a gunshot is “heard” it can be quickly integrated with video and imaging feeds to the spot where this occurred. As in factories, where we need to prevent workers from being injured or worse, low-latency real-time networking can immediately notify public safety agencies about an incident while prioritizing multiple parallel data streams—e.g. separating the sounds of birds chirping from the detected gunshot.

To download a PDF version of this essay, click here. ?

To download our Technology Insight “The Advent of Hybrid Networks,” go here and fill out the form at the bottom of the page. ↓