[Xenomai] Odp: Re: RtNet and Interrupts

Mariusz Janiak mariuszjaniak at wp.pl
Fri Feb 17 12:47:22 CET 2017

Dnia Piątek, 17 Lutego 2017 12:19 Norbert Lange <nolange79 at gmail.com> napisał(a) 
> 2017-02-17 11:45 GMT+01:00 Mariusz Janiak <mariuszjaniak at wp.pl>:
> > Dnia Piątek, 17 Lutego 2017 11:18 Norbert Lange <nolange79 at gmail.com> napisał(a)
> >> Hello,
> >>
> >> I wonder if the Rtnet Stack is driven at packet level, ie. if after
> >> every one received packet( or even buffer) an IRQ is received.
> >> I worked on a Project where a similar Problem with Ethernet
> >> communication had to be solved, and I just queued up enough receive
> >> buffers and just picked up anything that arrived at the start of a
> >> timeslot.
> >>
> >> On 100Mbit and a 1ms timeslot, I calculated a max. of 150 received
> >> pakets. Not that realtime communication can get anywhere close to
> >> that, but you have to deal with worst cases and guarantee that your
> >> system aint missing the cpu time spend in IRQs for other things - even
> >> if you just decide to drop all those packets, the IRQ overhead would
> >> be there.
> >>
> >> So is Rtnet using "polling" at timeslots, or irqs?
> >
> > Hi,
> >
> > As far as I know, RTnet is interrupt driven stack, it means that after reception of each packet the interrupt is generated and handled by the NIC driver. This is one of the reasons why porting of the RT driver  to new NICs is so problematic, especially now when new drivers support NAPI.
> Seems to be a bad Idea, I guess most RT communication is continuous,
> means there is not a question whether packets arrived but only how
> much of them.If I worked with irqs in the project above, the weak ARM
> CPU wouldn`t be able to do anything else.

Maybe author of the RTnet stack can give better answer, but in my opinion, in the case of the hard RT communication, we have to know in advance what is the maximum packet delivery time. The RTnet guarantee fixed time packet delivery. There is always a trad off between bandwidth and latency, you can't have both. 

> No efforts in using something similar to "New API"? (who comes up with
> such names...)

I remember a discussion about NAPI in context of  RTnet stack, but I don't know if any attempts have been made. 

> >
> >> Kind Regards,
> >> Norbert
> >>
> >> PS.
> >> Complete unrelated, but are there any big companies using Xenomai /
> >> Rtnet that can be named?
> >
> > I know that Universities use it a lot in the field of robotics (eg. SOEM + OROCOS). There is a port of the RTnet stack for microcontrolers if you are interested (RTnet Embedded)
> >
> > https://bitbucket.org/mariuszjaniak/rtnet-embedded
> Actually I am, sounds like you know about "RTnet Embedded" =).
> The reason I ask is that I want to push this at the company I work,
> unfortunately higher ups typically are most impressed if some rich
> companies are using it (and claims that it is the most popular thing
> ever).

Unfortunately I can't help. 

> >
> > Best regards,
> > Mariusz
> Thanks, Norbert

More information about the Xenomai mailing list