On Mon, 27 Apr 2026 03:14:55 +0200,
David Gwynne <david@gwynne.id.au> wrote:
> 
> > 
> > > > Numbers for iperf in single thread when two machines are in different vlan
> > > > but on the same cnmac:
> > > 
> > > I think multi-queue processing should be benchmarked with multiple
> > > flows. Also, I believe it is more common to forward traffic between
> > > different ports.
> 
> moving packets between ports is common cos it's often easier to set up.
> a "router on a stick" using a single port with multiple vlans is also
> common and valid, just maybe less common than a router with multiple
> ports.
>

What depends on hardware. Juniper SRX300 for example seem to have just one
cnmac interface which is linked to embeded switch. Device has 6 rj45 plusg
(plus two sfp) on one cnmac.

Is it a router with multiple ports? Yes, by look but no from OS point of view.

> > 
> > Sure, and this was just a naive benchmark to explain why I've used vlan in
> > tag in the first place.
> > 
> > On my ER4 with vlan and pf, when traffic is routed from one vlan to antoher
> > vlan on the same cnmac I do have ~923 mbit/s with 4 streams of iperf.
> > 
> > Here a cumulative diff (LRO + multiple queue) which address all your remakrs
> > and excluded already commited parts by kn@ and me.
> 
> can cnmac do toeplitz hashing for picking which rx thing to deliver a
> packet to?
> 

From fast reading it can be done as sofrware implementation, and I don't see
how it can be done by using hardware, at least for now.

-- 
wbr, Kirill