Page 23

Industrial Ethernet Book 105

The implementations in the magenta and aqua planes use OPC UA Pub/Sub over raw Ethernet with frame aggregation. However, potentially using Pub/Sub over UDP/IP shows an indistinguishable plane, while potentially using single frames increases the cycle times for payloads over approximately 50 bytes. Figure E shows that an advantageous implementation of OPC UA TSN with Gigabit physical layer outperforms existing solutions (based on 100M bit) by approximately a factor of 18. *) Profinet IRT cycle times are always multiples of 31.25 μs †) The ridges in the cycle time plane represent the use of a new Ethernet frame Cycle time comparison Over the years, the tendency has been to compare Industrial Ethernet technologies based on their respective feature sets. Even more important, however particularly important in motion control applications is the performance of the technology, measured in terms of smallest cycle time that can be achieved for a particular application. It can be seen as the most challenging metric, and if a technology fulfills this requirement, it can also be utilized in less timely challenging environments. The smallest achievable cycle time is the time required for a PLC to send all outputs to its slaves and receive all of their inputs in return. It is important that all slaves receive their outputs from the PLC within the same cycle. The first component of the cycle time is the link transmission delay. This refers to the time needed to send all frames over one wire with a specific link capacity. The basic equation for a summation frame is: τ = 8(header+max(remainder,n × (x + subheader))) C The remainder is the number of bytes needed to fill a minimum-sized Ethernet frame (84 bytes including inter-frame gap). For EC specifically, the formula translates to: τ = 8(40 + max(44, n × (x + 12))) C It should be noted that this formula considers only one frame. If the maximum Ethernet frame size is not sufficient, at least one more minimum-sized frame must be sent. Additionally, since device sub-payloads cannot be divided across multiple frames, the maximum Ethernet frame size will not be reached, and the data will have to be sent in the second (third...) frame. The second component of the cycle time is the propagation delay of frames through the network infrastructure, including the wires. For EC, the frame is sent through the entire network and back, resulting in a minimum cycle time of: 4.2018 industrial ethernet book Γ = (2n − 1)l + 2nδ + τ For PN, one must consider individual frames per node, rendering per frame: τ = 8(38+max(46,6+x)) It will be assumed that the frames are scheduled to arrive at the PLC, and the frame of the first slave passes one infrastructure device plus one cable. Γ = δ + l + n × τ All equations introduced here assume simplistic cases, where input and output data volumes are equal and the topology is a perfect line. In real applications, the comparison depends on many additional parameters: • Ratio of input data to output data • Percentage of devices with direct cross traffic • Utilization of different cycle times • Topology (line, star, ring), and hence number of hops between devices • Availability of modular I/Os with backplane bus Results assuming more realistic values are shown in Figures A and B (using 100 M bit). Using a different link capacity (1 Gbit) changes the situation quite dramatically, since only the transmission delay component of the cycle time, and not the network infrastructure component, can be reduced by a factor of 10. Hence, the performance of technologies with a larger dependence on infrastructure (EtherCAT, Sercos III, POWERLINK) improves on average by a factor of 4 - 6 when using Gigabit. In contrast, technologies based on switched Ethernet (EtherNet/IP, Profinet IRT) can leverage a factor of 7 -10 for large enough payloads. For small payloads, the transmission delay of a short frame might be smaller than the infrastructure latency, resulting in a lower bound for the minimum cycle time in a line. Today’s COTS cut-through switches for Gbit have forwarding latencies in the range of 2 μs, which translates to a minimum frame size of 250 bytes (=2000 bits) (neglecting propagation delay on the wire). Sending smaller frames does not further decrease the cycle time. In applications with demanding performance requirements, devices with short forwarding latencies are crucial. The calculation of the cycle time for OPC Comparison of minimum cycle times of OPC UA TSN @1Gbit w/ existing technology Figure E dark orange: Profinet IRT*); light orange: EtherCAT†); red: POWERLINK; magenta: OPC UA TSN


Industrial Ethernet Book 105
To see the actual publication please follow the link above