On Mon, Apr 17, 2006 at 01:05:29AM -0400, Lee Revell wrote:
> On Sun, 2006-04-16 at 23:22 +0200, torbenh@email-addr-hidden wrote:
> >
> > Work is underway in improving the latency for big channel counts,
> > like 24in / 24out. There seems to be a bottleneck in the kernels
> > handling of big UDP Packets.
>
> Nice, I'm looking forward to the LKML thread!
actually i dont want to report it.
i evaluated "the irc opinion" and came to the conclusion:
- most people think udp is evil ;)
- it looks like i am the only one messing with BIG (16k) udp packets in
low latency situations under high SCHED_FIFO load.
i am currently implementing packet fragmentation into netjack.
if that improves the situation drastically i would start reporting on
this. in fact i am pretty annoyed by the kernel.
i would be very happy, if somebody (sletz ?) could test the performance
of 24ch connections under OSX.
and i dont have an rt-kernel on the second machine so i cant verify if
its already fixed in rt-preempt. Perhaps we need some kind of rt-network
packet.
/me hides.
>
> (you do plan to report it right...?)
not really :/
-- torben Hohn http://galan.sourceforge.net -- The graphical Audio languageReceived on Mon Apr 17 16:15:06 2006
This archive was generated by hypermail 2.1.8 : Mon Apr 17 2006 - 16:15:07 EEST