Real-time Ethernet for industrial control

Search for the word ‘latency’ within the IEEE 802.3 CSMA/CD, known more commonly as Ethernet, and there will not be many occurrences. Why? Because Ethernet was never originally designed for latency-critical, real-time networks, such as industrial control, voice and video.

Today’s incredibly low Small-Office and Home-Office (SOHO) prices, combined with interoperability and openness, have driven Ethernet from the office into our homes and factories. Designing Ethernet into such applications needs careful consideration and an understanding of the architecture of switches, repeaters and transceivers to ensure success.

 

Switches and repeaters

All of today’s Ethernet switches are based upon store and forward architecture, which stores the complete packet and processes prior to forwarding. By analysing the complete packet a host of features are now possible, such as Virtual LANs (VLANs), priority classification and error checking. The consequence of store and forward architecture is, however, a non-deterministic latency behaviour. This is demonstrated by the latency measurements in Table 1 below using the Micrel 5-Port Switch, KSZ8995MA, as an example.

The total latency is derived from two components, packet size and the internal forwarding delay, and can be calculated as below:

Total Latency = (Packet size x 8) / rate + Forwarding delay.

Table 1 shows a constant and small forwarding delay of around 2.7µS for the Micrel switch, independent of packet size. Hence, fixing the size of the packets in a network will provide constant switch latency. To reduce the overall switch latency the packet size should be reduced to a minimum. However, in reality this may be difficult to control in a network.

 

 

Repeater hubs result in a much simpler architecture than switches. Repeaters deploy a cut-through architecture, which significantly diminishes latency by forwarding the incoming packet to all ports, with the exception of the ingress port, before receiving the complete packet. This technique offers less in terms of features but delivers improved latency and packet-size independence compared to the switch counterpart.

 

Industrial control networks

Switch and repeater latency is of major significance when deploying Ethernet into the unique ring topology of industrial control networks (typically star topology is found elsewhere). Here, total network latency equals the sum of each of the individual node (Ethernet switch or repeater) latencies. Solutions for time-critical control networks have led to a number of industrial-Ethernet standards evolving, such as PROFInet, EtherCAT, EtherNet/IP, Modbus and Powerlink. All adopt Ethernet as the physical layer but utilise Layers 3 and up of the Open Systems Interconnection (OSI) network model to deliver data with accuracies to within 1µS.

 

 

Real-time performance is generally implemented by time multiplexing the communication on the network between the different nodes. Critical isochronous data transfer is performed at the start of the cycle, followed by non-critical Transmission Control Protocol/Internet Protocol (TCP/IP) traffic. Such methods guarantee the real-time performance of the network whilst still offering compatibility with office networks. This method is demonstrated in Figure 1, using Ethernet Powerlink as an example. The Powerlink software stack controls the data flow on the network using a method called Slot Communication Network Management (SCNM). One station, the Powerlink manager, assigns timeslots for each of the stations in the network (cyclic period, Figure 1). Stations are only allowed to send data when requested by the manager. As a consequence, only one station can access the bus at any particular time, removing the possibility of collisions to achieve deterministic performance. At the end of the cyclic period the network is free to deliver any non-real-time TCP/IP traffic (asynchronous period, Figure 1). This provides compatibility with the office layer of the network.

To reduce latency jitter in the network, the Ethernet Powerlink group recommends using 100Base-TX/FX Ethernet repeater hubs. However, most Ethernet repeaters today are obsolete in favour of intelligent switches. The scarcity of Ethernet repeaters often leads to FPGA-based implementations with external Ethernet PHYs, increasing cost, risk and time-to-market

This trend is bucked by Micrel’s KSZ88xx, a new generation of switches and controllers offering a unique low-latency repeater mode. This mode provides a maximum port-to-port latency of 310nS and a total deviation of 40nS, making it ideal for real-time critical applications.

 

Higher-level protocols

An alternative approach to delivering real-time packets is provided with higher-level protocols such as IEEE 1588. IEEE 1588 utilises User Datagram Protocol (UDP) packets over IP on the Ethernet network to provide synchronisation down to a microsecond. Such performance meets the stringent real-time requirements for motion control applications. ProfiNet, EtherNet/IP and the EtherCAT industrial- Ethernet groups have all adopted this standard for network synchronisation. Figure 2 shows a typical hardware implementation of the IEEE 1588 functionality.

 

 

Network synchronisation is achieved using the Best Master Clock Algorithm (BMCA) to select timing. If the node is to be a master to many of the nodes in the ring then a highprecision source is used, such as the Global Positioning System (GPS) device shown in Figure 2. If the node is not designated as a master then it will extract timing from the network using the IEEE 1588 protocol or, failing that, the on-board local oscillator.

To synchronise the master and slaves, IEEE 1588 operates Precise Time Protocol (PTP) based on IP multicasting. The synchronisation process is divided into two phases as shown in Figure 3. First the offset time between the master and slave is calculated and corrected.

 

 

To perform this function the master continuously transmits a unique message to the slave at defined intervals, usually every 2 seconds. The second phase of the synchronisation process is the delay measurement. The slave will send a delay request to the master, which is returned and the round trip delay calculated using timestamps. The assumption here is that the delay between master and slave is always symmetrical.

 

Conclusion

The accuracy of the time synchronisation will depend greatly on the ability to time stamp packets as close as possible to the physical-layer line interface, eliminating any switching- or software-variable delays. Today hardware implementations are typically realised in an FPGA, and can provide network synchronisation to within less than 1µS. Alternatively an IEEE 1588 software-based solution can be implemented with time stamping performed by the processor. However, it is difficult to compensate for the non-deterministic latency of the switch, resulting in synchronisation accuracy of around 10µS to 100µS. Consequently although providing a real-time solution, this approach is unsuitable for precision applications such as motion control.

As Ethernet broadens its appeal from the factory to the home, it is facing challenges that were inconceivable when the IEEE 802.3 specification was first published back in 1985. Next-generation silicon will need to provide further enhancements to meet the demands of real-time applications. For now, careful consideration and an understanding of the Ethernet architecture is essential when designing such products.

 

www.micrel.com/page.do?page=productinfo/ether_over.jsp