> 22.02.2026 18:25, Kirill A. Korinsky пишет:
> > I've tried it on my EdgeRouter 4.
> > That I do have with snapshot which I just installed:
> > machine 1 -> ER4:
> >
> > [ ID] Interval           Transfer     Bitrate         Retr
> > [  5]   0.00-30.00  sec  1.39 GBytes   397 Mbits/sec  15928             
> > sender
> > [  5]   0.00-30.01  sec  1.39 GBytes   397 Mbits/sec                  
> > receiver
> > machine 1 -> ER4:
> >
> > [ ID] Interval           Transfer     Bitrate         Retr
> > [  5]   0.00-30.00  sec   590 MBytes   165 Mbits/sec  65568             
> > sender
> > [  5]   0.00-30.06  sec   590 MBytes   165 Mbits/sec                  
> > receiver

> Thanks for testing, I certainly did not expect such a massive drop.
> I'll test here as well and try to see what this is about before committing 
> anything.
> I wonder, if this is "just" a weak CPU or something else, e.g. specific to
> octeon, NIC drivers and/or in your setup.
> Can you share your iperf commands and a dmesg for completeness, please?
>
> Cc'ing Janne:  You tried a similar diff on your boxes;  care to share
> which ones and whether you saw any noticable change in performance?

My tests also included a diff to make cnmac(4) put ifq's to different
softnets which it doesn't now, since the ifq placement uses the softc
"index" and at least on my ER-8, all eight cnmacs have index 0, but
different "ports" or some other property in softc, so I also had a
silly counter bumping the number by one for each attaching cnmac to
make them end up on different ifq -> softnet threads, otherwise I got
ncpusfound softnets but all cnmac always ended up on softnet0.

I will rebenchmark on erl3 and ER8 to see how they react, but as
noticed, just having more threads should not really hurt like this,
worst case should have been "some slight overhead but no improvements"

-- 
May the most significant bit of your life be positive.

Reply via email to