原创 Deterministic networking to go mainstream?

Before the Internet began turning into a means of distributing mass entertainment and infotainment, there was once a clear delineation between two types of network operation.

One type included the real time and deterministic networks that served the specialised needs of industrial/ machine control and factory automation. The other included asynchronous networks in business and personal computing where the need was for reliable delivery and for which determinism and precisely timed delivery was a secondary consideration.

This is changing rapidly. Many of the early factory automation and machine control oriented networks operated according to the same rules of real time and deterministic operation of the devices they linked. Starting as ad hoc and mainly proprietary solutions, out of this came a variety of so-called controller area networks standards such as CANopen, ControlNet, DeviceNet, Modbus, Profibus, and the fieldbus.

At the same time, network communications between non-real time systems in business and data processing began to standardise around such LAN-based topologies such as Ethernet, which used something very similar to the current TCP/IP internetworking protocol suite.

Unlike the controller area networks, the TCP/IP protocol is probabilistic and asynchronous in nature. It was—and by in large still remains—a network environment where the aim is to guarantee delivery, but not within any particular, or even predictable, time frame. Even with the development of enhancements such as the Network Time Protocol (NTP) for clock synchronisation between computer systems, it is certainly not within the fine-grained microsecond and millisecond deterministic boundary conditions most embedded control operations need.

But Ethernet and the TCP/IP suite have one advantage that these real time network protocols did not: they are ubiquitous, which provided two things that embedded developers needed: 1) a network protocol that if engineered correctly would work in a variety of environments and allow communications between them, and 2) a protocol that was low in cost to implement because of its ubiquity and the resultant economies of scale.

But starting in the late 1990s and through most of the last decade there have been increasing efforts to adapt the Ethernet protocol to the real time and deterministic requirements of embedded control applications.

Initially it was just a matter of physically configuring stations or nodes into a closed Ethernet collection and limit the number of nodes until the response times were within the deterministic requirements needed. This way the time it took for each to do the time consuming handshaking to ensure delivery could be minimised and, more importantly, predicted.

Even more fine-grained were early attempts to take the underlying TCP/IP protocol apart and use only those portions, such as UDP—that do away with the time consuming process of acknowledging receipt to guarantee delivery—in a closed environment to handle transmissions in a much more deterministic way.
But most useful to embedded developers was the creation in 2002 of the IEEE 1588 and its precision clock synchronisation protocol. Using it as the starting point, many of the industrial control network protocols have adapted elements of the PTP and out of that has emerged what is called the Industrial Ethernet.

Joining Ethernet in adapting to a networking environment that requires more and more real-time deterministic performance have been any number of other specifications and standards. One example is the Firewire serial interconnect specification which started its life as an alternative to the Universal Serial Bus as a way of networking various peripheral and storage functions to a main server or desktop computer.

Such adaptations to real time and determinism are coming just in time for the broader Internet, which is now increasing being used as the means by which to deliver real time video transmissions to Internet-enabled TVs, smartphones and tablets. For example, virtually every significant Ethernet router/switch manufacturer has incorporated the IEEE 1588 standard into their hardware layer.

Also, the 1588-based IEEE 802.1 Audio/Video Bridging enhancement to the Ethernet standard is being widely used to deliver time-synchronised, low-latency audio and video while retaining 100% compatibility with Ethernet.

Further contributing to this trend is Introduction to the Synchronised Ethernet (SyncE).Taking a slightly different approach to making the TCP/IP suite more deterministic, this ITU-T standard facilitates the transference of clock signals over the Ethernet physical layer. It is being used widely in applications such as cellular networks, IPTV and VoIP, not to mention such Internet-backbone access technologies as passive optical networks which require something more than traditional Ethernet protocols.

Also not immune to the need for a more deterministic TCP/IP protocol is the new segment of embedded design activity involving in bringing wirelessly connected M2M and Internet-of-Things enabled devices into homes and buildings via new IPv6 Internet protocol extensions such as 6LoWPAN. Ironically, one way developers are looking to satisfy this need for more deterministic operation is the same UDP subset that so occupied embedded developers attention in the late 1990s.

I do not see this trend towards the incorporation of real time determinism into the broader Internet slowing down. Soon, the deterministic modifications to Ethernet servers and routers that are the backbone of the Internet will move out of the physical layer where they now reside and move into the protocol layer. This trend will be driven not by how humans use the network, but by the needs of the many more embedded and distributed sensors that will be connected.

Where present protocol standards now reflect the response times of humans (one to several seconds), the Internet of the future will be mainly populated and used by devices with response times in the microseconds to the milliseconds. Even now, in the average home, the ratio of devices to humans is on the order of 10:1. In the near future it is virtually certain that most of the devices will be connected.

We live in exciting times. After graduating from Columbia University in the 1970s, I came into the electronics industry only a few years after the introduction of the microprocessor. In each of the decades since. I thought the one I was in was the most exciting ever and that there was nothing that would top it. This decade is no different and has not disappointed me yet. I can't wait to see what the next few years in this new exciting segment of the industry will bring.