In this article a comparison in latencies of two COTS IEC 62439-3 switch IP cores implementable on FPGAs is shown. The first one, combines Cut-Through with Store-and-Forward switching architectures, and the second one is based only on Store-and-Forward switching technique.
Reliable Ethernet Networks – HSR & PRP
Reliable Ethernet Networks are gaining acceptance for many Industrial Automation applications. One illustrative example of this evolution, is the adoption by the International Electrotechnical Commission of the High-availability Seamless Redundancy (HSR) Ethernet based protocol and Parallel Redundancy Protocol (PRP) for Power Substations Automation (IEC 62439-3 clauses 5 and 4). Both protocols offer zero switchover delay time, no-frames lost in case of failure, and strong means for Network Supervision at Layer 2.
HSR frames are not compatible with conventional Ethernet infrastructure. However, PRP allows sending duplicate frames through two conventional Ethernet Networks. Thus, the PRP field of application is very wide although it has not been specifically designed for ‘Real-Time’ Ethernet Environment.
‘Real-Time’ implies answering signals at a predictable time after their occurrence. For instance, modern digital control loops need reaction times lower than 10µs. Latest Ethernet-based control protocols, like Ethercat or Sercos III, are based on hardware implementations that offer predictable behaviors for synchronization and very low latency times.
- HSR has been designed to meet the strict communications requirements set for Process Bus. HSR interconnects Intelligent Electronic Devices (IEDs) in each bay.
- PRP is suitable to be used for Station and Inter-Bay Buses. Thanks to its flexibility, it can attach many heterogeneous devices.
To maintain redundancy in the communications, the interconnection between PRP and HSR networks is performed using redundant gateways. Each HSR link is connected to each PRP LAN using two gateway devices. Thus, a potential ‘Single-Point- of-Failure’ is avoided.
Cut-Through vs Store-and-Forward
Both, Cut-Through and Store-and-Forward L2 switches base forward decisions on the destination MAC address of the packet. The main difference between them is:
- Store-and-Forward switch makes the decision after the whole packet has been received.
- Cut-Through switch makes a forwarding decision after analyzing the destination MAC address, which is in the first part of the frame.
In a Store-and-Forward switch, the latency time includes the time needed to receive the whole frame. Thus, the latency times are worse in comparison to Cut-Through switches.
Forwarding Latency times
Implementing both approaches (one mixing Cut-through and S&F and the other one a pure S&F) in our HSR-PRP IP Core on a Xilinx FPGA, the results are as follows:
The HSR-PRP IP Core implements a switching architecture specifically designed for PRP and HSR protocols.
The theoretical minimum latency time is calculated by taking into account the mandatory fields of the Ethernet frames that must be analyzed to make the switching decision. In that case, in Cut-Through the time is independent of the frame length, so it will be constant.
In the case of a pure Store-and-Forward approach, the whole frame needs to be stored before starting the re-transmission, so the latency is dependent on the frame length. It can be seen that it is one order of magnitude greater than optimized Cut-Through switching architectures.
The analysis shows that a custom architecture that combines Cut-Through with Store-and-Forward approaches offers the best latency times under any circumstance.
The role of FPGAs in these new protocols is vital. On one hand, they allow low-latency, flexible and scalable solutions for strict requirements set in these standards. On the other hand, they reduce the time-to-market and risk for industrial manufacturers when they provide equipment for new and specific protocols and markets.
For more information on this study, please contact us at: firstname.lastname@example.org