[etherlab-dev] 1 datagram UNMATCHED! / working counter changes

[Daniel Bechter]

Hey guys,

I am currently performing some realtime tests with a PowerPC 1.5 GHz. Loop
is running at 10 kHz (100us). Here is my EtherCAT topology:

PowerPC -> EK1100 -> EL4732 -> EL4732 -> EL4732

The logical process image comprises 12 Byte only (4 Byte per analog out
slave).

Problem now is, that  I get error messages from the domain approx. every
second (I am checking the domain state with every cycle, so every 100us):

EtherCAT WARNING: 1 datagram UNMATCHED!
EtherCAT: Domain 0: 2 working counter changes - now 3/3.
Domain1: WC 0.
Domain1: State 0.
Domain1: WC 3.
Domain1: State 2.

I also sniffed the Ethernet packets with an ET2200 but could not find any
errors. A packet with LWR datagram leaves every 100us (with WC 0) and about
2us later it returns after being proccessed by all slaves (with WC 3).

At 5 kHz and 1 kHz, the error comes up with the same periodicity (and always
1 unmatched datagram with 2 working counter changes). At 100 Hz, everything
is fine.
Something similar happens on a 3 GHz P4, but much later, 5 kHz is almost
stable and at 10 kHz the same error starts showing up, however with more
than 1 unmatched datagrams and more than 2 working counter changes per error
(I am wondering how this is possible if I check the domain every cycle...)..
What makes me wonder is that if I put some load on the P4 system, it also
seems to fail to adhere to the 5 kHz loop with the above mentioned error.

What is going wrong here? I am not sure whether it is really the PC that is
too slow, as I can see the packets are sent and must have been received with
all slaves having done their job (judging from the WC).

The application runs in user-space without any realtime patches applied
(kernel 2.6.26.8). I have done some realtime tests before by sending an
Ethernet frame every 100us via raw sockets, also from a user-space
application. Jitter was 1-2us.

Does anybody have any comments regarding that?

Greets,
Daniel