DDS over TSN: Future-proofed architectural framework for industrial automation

Feature

8 December 2023

In an age where data is at the core of every decision-making process, “data” and “connectivity” are the base of any modern automation system. On the one hand, they open up a new world of possibilities like autonomous cars and smart cities, but on the other, the increasing number of connected devices and the sheer volume of data can burden and even overwhelm networks.

This dilemma is driving new ways to store, analyse and process data closer to the source – i.e., Edge computing. Edge computing addresses the physical limitations of networks, notably bandwidth, congestion and latency, by decentralising the network. Simply reducing the distance data needs to travel eliminates latency, whilst reducing the load on the network’s bandwidth eliminates congestion.

Modern protocols like MQTT and OPC UA pub/sub are becoming popular technologies that bring new applications closer to the edge. They have been designed (or even retro-fitted) as lightweight publish/subscribe messaging transports. Further, they are ideal for connecting remote devices to edge applications or quickly integrating devices into centralised cloud infrastructure.

However, these solutions are not perfect; what’s needed is a single approach that handles both edge and cloud connectivity in an interoperable manner.

Data-centricity and convergence

In industrial settings, raw data consists of individual facts that lack context, are devoid of meaning and difficult to interpret. However, information can best be described as a set of data in context with relevance to one or more things at a point in time or for a period of time – thus, information must have both relevance and a time frame.

Data has a certain value, and understanding that value requires a methodology for data valuation. Hence, system architects are moving to data centricity for data valuation. Data centricity is an architectural pattern where data is the primary and permanent asset, regardless of the application. A data-centric architecture uses unified data models to describe a system in terms of the information exchanged, not devices or applications. Data models provide schemas (what information is flowing and how it is related) and a control model (how and when it flows).

Also worth mentioning is that the data produced today has evolved from simple time-series type (timestamp, key, value) to including advanced sensor data, real-time video streams, LIDAR, and real-time GPS location data, among many others. Managing these complex types of data requires a converged infrastructure solution to provide real-time capabilities and simplify data flow management within a single data-centric connectivity infrastructure.

DDS for data-centricity

The Object Management Group (OMG) Data Distribution Service (DDS) standard is a platform-independent software framework used to architect and implement data-centric software. A DDS data model, a relational concept like a database table, provides a schema, while DDS Quality of Service (QoS) provides exact control over data rates, deadlines and other data-flow parameters. By defining a shared, secure “global dataspace”, DDS provides a “single source of truth”, enabling a high degree of cohesion (how related the functions within a single module are), whilst guaranteeing loose coupling (the interdependencies between modules).

Distributed systems express coupling in four dimensions: time, space, flow and types:

Time – No dependency on startup or join sequence. Participants may come and go; adding or removing applications or flow paths doesn’t affect others.
Space – Data can come from any physical location and from any producer. Producers and consumers may reside in devices or applications anywhere. In a larger system, applications can transparently live at the edge, in the fog, or in the cloud.
Flow – Data flow rates or reliability specifications between endpoints do not interact. Each application can request data at a different update rate, over any network, and with or without reliability guarantees.
Type – Automatically converts dissimilar types for a data flow if they “match”, allowing systems to evolve over time.

DDS makes it seem like all the data in the system is local. Applications read and write to a “global dataspace” that looks like local memory, and the data-centric DDS middleware ensures it contains the right data. Unlike OPC UA, OPC UA pub/sub and MQTT, there are no clients, servers or brokers. The global data space sits between every participant providing secure, deterministic access to information without tight coupling.

The most important capability of DDS is the Quality of Service (QoS). DDS allows every application to request 21 different parameters such as deadlines, latency budgets, update frequencies, history, liveliness detection, reliability, ordering, filtering, and more. These QoS parameters allow system designers to construct a distributed application based on the requirements for, and availability of, each specific piece of data. Examples include:

Durability allows late-joiners to get data that was produced before they started.
Deadline and separation specify minimum and maximum data update rates for each subscriber.
Liveliness ensures that each dataflow is healthy.
latency budget, transport priority and reliability decouple flow on a per-stream basis.

Convergence to the architecture

Ethernet (standardised under IEEE 802.3) is one of the original networking technologies. Because of its simplicity to deploy and its ability to evolve without losing backwards compatibility, it has become the de facto standard in IT networking. Despite being around for nearly half a century, it was only in the last decade that the OT side of the house started to incorporate Ethernet into their solutions.

Industrial applications usually have strict temporal and deterministic requirements. By definition, Ethernet doesn’t guarantee deterministic message delivery or real-time behaviour. However, extremely high performance enables it to serve most such applications, provided there is a way to manage network traffic.

Two factors which have slowed the adoption of Ethernet in the industrial space are jitter, the variation in the delay between packets arriving, and latency, the time it takes a packet to reach its destination. Since the original Ethernet specification lacked determinism, it was believed that the protocol could not be used reliably in many machine applications where this lack of determinism could lead to poor quality or even machine damage.

Time-Sensitive Networking (TSN) is solving the issue of standardised real-time communication based on standard communication hardware. TSN is an extension of standard Ethernet that regulates the data communication in Layer 2; see Figure 1.

It’s important to remember that TSN is only a “pipe” for getting data from one place to another in a deterministic way. It does not address higher level application functions, such as safety or motion control. The real value of TSN is the ability for OT to virtualise industrial networks the same way many organisations have consolidated their server infrastructure in the cloud. Enabling multiple types of traffic to share the same networking hardware provides the basis of convergence.

DDS over TSN

Together, DDS and TSN lead to a converged approach for handling real-time communications in industrial applications. DDS, by design, is network transport agnostic. DDS QoS enables deterministic applications to be built regardless of the transport. DDS relies on the underlying network transport for hard real-time requirements. TSN makes standard Ethernet real-time and enables the convergence of multiple kinds of traffic on the same network (virtualisation). DDS provides a deterministic, platform-independent data-centric framework for architecting and implementing highly distributed systems from the edge to the cloud.

DDS over TSN (Figure 2) is the most interesting development in industrial communications since the development of Ethernet. The combination of these two standards brings real-time, virtualised (converged) data centricity to industrial automation. The challenge of turning raw data into actionable information that can be acted upon in real time is now possible, without the inherent architectural limitations of MQTT, OPC UA and OPC UA pub/sub.