Re: [RTnet-developers] High performance Giga Ethernet possible with RT-net?

[Jan K. <jan...@we...>]

Fillod Stephane wrote:
> ...
> I'll stop here my random thoughts for now, because I'd like to know
> whether
> RT-net is the right place. Besides, some people may have already taken
> that
> route, or at least that goal, and why not escape this wheel of
> reincarnation?

Sorry for cutting down your _very_ interesting mail that much, but I'm
short on time. A few comments that shall not end this discussion by not
citing your individual ideas, rather let the ideas flow even more:

 - RTnet is not only RTmac/TDMA. Just think of EML (Ethercat Master Lib)
   over RTnet, it has quite similar requirements! Also, several users
   are running plain RTnet/UDP against "black-box" devices without TDMA.

 - Zero-copy RX is on my list as well, probably with single-user mode
   for the beginning until HW-demuxing becomes commodity. There are also
   improvements for certain scenarios along the wakeup path possible.
   Same for zero-copy TX.

 - New APIs should always consider what happens around "normal" Linux
   first. Only divert from (upcoming) standard approaches when hard RT
   demands it.

 - NIC polling instead of IRQs: Sounds like Linux's NAPI. The drivers
   should look very similar under RTnet, one only has to reconsider the
   mode switch (IRQ-driver <-> polled) policy /wrt determinism.

 - There are certainly some things to do to push RTnet in the sketched
   directions, but we should be able to do this evolutionary, step by
   step.

Jan