This is basically a rewrite of the simple_test program included in the SOEM project, meant to highlight the differences between SOEM and IgH-EtherCAT.

Major differences from simple_test:

• this is a self contained project;
• the threaded code has been removed;
• the SOEM version uses the new APIs intead of the legacy ones;
• by default it updates (every 5000 us, customizable in the arguments) a binary counter on the first 8 digital outputs; if you explicitely set the period to 0, it measures the roundtrip performances.

Two implementations are provided: ethercatest-soem, based on SOEM, and ethercatest-igh, based on the IgH-EtherCAT software stack.

The two projects are really different.

SOEM is basically a userspace library. It is really easy use: you just need to link your application to it (statically, by default) and you are ready to go. Thr program tries to guess the Ethernet device to use but you can override its logic by passing it as first argument. To be able to access the raw Ethernet device, the SOEM program must be run as root.

The IgH EtherCAT stack instead is much more complex. A couple of kernel modules must be configured and loaded before: ec_master and an EtherCAT device. In these tests, the generic EtherCAT device (ec_generic) was used. Furthermore, the application must be linked to a userspace library. The user running the program must be able to access the EtherCAT device created by the IgH modules, typically /dev/EtherCAT0. This is usually owned by root but, by leveraging the udev infrastructure, you can change ownership and mode to be able to access it from a normal user.

The master hardware is a consumer PC with an i7-7700T CPU at 2.90 GHz. The NIC interface is an e1000e Intel-based one operating at 100 MBit/s. A quite recent linux kernel (5.11.2) was used.

The PC is connected with a 2 m category 5 UTP cable to a Beckhoff node with three slaves: one EK1100, one EL1809 and one EL2808.

These are the results of ethercatest-soem 0 on an idle system:

Roundtrip time (usec): min 104  max 599

Output on the same system while building a project in another console:

Roundtrip time (usec): min 76  max 7372

# Using `sudo nice -n20 ethercatest-soem 0`
Roundtrip time (usec): min 77  max 39758

# Using `sudo nice -n-20 ethercatest-soem 0`
Roundtrip time (usec): min 76  max 3031

The performances are not stable, meaning that subsequent calls can give quite different results, e.g.:

Roundtrip time (usec): min 101  max 661
Roundtrip time (usec): min 97  max 680
Roundtrip time (usec): min 93  max 698
Roundtrip time (usec): min 78  max 729
Roundtrip time (usec): min 190  max 731
Roundtrip time (usec): min 169  max 590

These are the results of ethercatest-igh 0 on an idle system:

Roundtrip time (usec): min 190  max 350

Output on the same system while building a project in another console:

Roundtrip time (usec): min 127  max 398
Roundtrip time (usec): min 122  max 405
Roundtrip time (usec): min 148  max 353
Roundtrip time (usec): min 121  max 382

In this case the performances are much more consistent and stable. By using a dedicated driver instead of ec_generic the results would likely be lower. The problem is the dedicated drivers must be kept in sync with the kernel, and this could become a maintenance nightmare.