[etherlab-users] Απ: AL status message 0x0036: "DC Sync0 Cycle Time"

[Mike Karam]

Well I think you are right, I don't think either that the Xenomai dance worths it. As for the driver, I'm not so sure, but I think it's not the only parameter to the equation.

Thanks again,
Mike

________________________________________
Από: Jürgen Walter • DATATRONiQ <jw at datatroniq.com>
Στάλθηκε: Δευτέρα, 2 Απριλίου 2018 10:22 μμ
Προς: Mike Karam
Κοιν.: etherlab-users at etherlab.org
Θέμα: Re: [etherlab-users] AL status message 0x0036: "DC Sync0 Cycle Time"

Hi Mike,


> I'm attaching the makefile I used. It's the makefile provided by
> synapticon: ...
cool - many thanks; this will certainly be helpful!!


> Can you answer to any of my questions? :)

>> I wanted to thank you all for your valuable advice. Now I'm acheiving
>> near 3.5 kHz loop rate without a single datagram loss. My configured
>> system is (just for reference): Ubuntu 16.04 with kernel
>> 4.8.15-rt10(preempt_rt patched) and support for native driver. I
>> would like to ask another question: In the dc_user example I don't
>> understand why the latency is measured that way. Could someone make
>> it more clear? This latency is the latency of our application (the
>> time the scheduler does to put our process to running, so that our
>> application can process the newly arrived datagram)? What quantity is
>> measured exactly with this latency? Bonus: Can we measure the time
>> the datagram does to leave and come back to the master, with
>> wireshark accurately?

I would love to  -however, only a user myself and not really deep into
the protocol stack (never even tried debugging the datagrams on the wire
(debug interface, Wireshark (but I want to!))

I will also share -once I get it all working to good satisfaction also
share my setup/steps to epiphany. fyi- I am not sure if it is really
worth doing the entire Xenomai dance - shouldn't latency (on a modern
and fast enough system (such as your laptop)) mostly be determined by
the quality of the ethernet driver and of course the chipset? Just
saying this, because I think the "preempt" seems pretty much standard
and is really simple to patch and compile.

Thanks!! Jürgen