Ultra Low Latency Network

No existing Industrial Network provided the performance (latency & throughput) that a customer application required, so we have developed this Ultra-Low Latency redundant network.

Introduction

Gigabit Ethernet over copper twisted pairs (1000BASE-T) is an ubiquitous, easy, and extremely reliable Physical Medium.
ALSE has developed a lot of experience around this technology, like the very innovative GEDEK communication kit (processor-less hardware stack) and allowed achieving reliably unprecedented data throughput. This technology has been used in hundreds of systems all over the world.

One of our clients expressed the need to build an architecture with up to more than 20 regulation and control boards interconnected together, and with a requirement that all boards should be updated with the regulation data from ALL OTHER boards during the first half of a single PWM cycle (targeted at 20 kHz). In other words, all the boards should see all the other board’s data in no more than 25 micro-seconds.

Another requirement was the ability to implement redundancy.

Quite naturally Gigabit Ethernet comes to mind to implement the inter-boards communication.
However, even under underestimated conditions (20 slaves, 20 kHz, no redundancy), 40 Ethernet frames must circulate in much less than 25 µs. This simple calculation shows that a standard (store-forward) architecture, even with a near zero delay (like GEDEK) would be outperformed by a ratio greater than four !

Are existing solutions available ?

We conducted a careful pre-study and reached the conclusion that NO SOLUTION existed in the market that reached only a fraction of the performance required !

The table below, extracted from a very interesting article about Industrial Ethernet shows the state of the art.

[table]

Industrial Ethernet standards Cycle Time

Our solution

  • We use completely standard IEEE 802.3-2008 Ethernet frames.
  • We implement a “circular” bi-directional (dual) ring. Two frames circulate in opposite directions for a full turn, thus already implementing a first redundancy.
  • We use the Cut-Through Forwarding technology in our own optimized switch to reach an absolute minimal latency.
  • We use the Summation Frame method (each node adds it’s own data in a slot of the circulating frame).
  • We use a fully Hardware implementation, thus removing any software interaction or limitation to the performance. In Hardware, decisions can be taken in less than 10 nanoseconds (ie practically at each incoming byte in the GbE link), with no jitter.
  • We re-used (and customized) our optimized and well proven Gigabit-only MAC (Media Access Control).

Note : We have implemented several clever optimizations to further reduce our cut-through switch latency. The details are only available after signing an NDA.

Network Topology

The figure below shows 9 boards connected together, one of which being the Master.

Network (ring) Topology

Test System

We have used 9 DE2-115 FPGA kits attached together mechanically to form a “tower”. See picture below. The DE2-115 (used in some Training courses) was available and it comes with two GBE interfaces (Marvell 88E1111 PHY).

This setup allows to practically measure with great precision the achieved latency in realistic conditions (including the delays in the Marvell Gigabit PHYs). The results were extremely close to our simulations.

Tower of DE2-115

Complementary tools

We have developed an “Embedded Sniffer” : a special version of the Master forwards all (or some) circulating frames to another (3rd) Ethernet port for analysis with a PC using Wireshark, without disturbing the Ring network.

Conclusion

Our technology (based on an optimized ultra-low latency Gigabit Ethernet cut-through switch) can enable a new range of industrial applications that were not feasible so far.

If your system consists of several boards that need to communicate with each other and exchange data with very small delays and latency, our technology may help you. Do not hesitate to Contact us

Back to top